Beruflich Dokumente
Kultur Dokumente
2018
Please cite this publication as:
OECD (2018) Rethinking Antitrust Tools for Multi-Sided Platforms
www.oecd.org/competition/rethinking-antitrust-tools-for-multi-sided-platforms.htm
This work is published under the responsibility of the Secretary-General of the OECD. The opinions expressed and arguments
employed herein do not necessarily reflect the official views of the OECD or of the governments of its member countries or
those of the European Union.
This document and any map included herein are without prejudice to the status or sovereignty over any territory, to the
delimitation of international frontiers and boundaries and to the name of any territory, city, or area.
© OECD 2018
FOREWORD │3
Foreword
Digitalisation of the economy has, it seems, arrived in waves. In the first wave, the
internet allowed us to buy directly digital copies and physical products and services from
online stores, rather than physical ones. The second wave has seen the appearance of
online platforms, which assemble, search, review and match users with sets of products
and sellers. To do so platforms recruit at least two, but often three or more sets of users,
many of which value the platform not for its own qualities, but for the presence of others
upon it. We currently await the third wave, said to involve the direct transfer of not just
information, and hence digital copies, but also of value, in the form of unique digital
products and services over the internet. As more and more physical products and services
become largely digital in nature, the scale of this next change becomes ever more
important. While the payment systems used by platforms may face challenges, and
platforms themselves may change in nature, they seem likely to remain crucial to our
ability to interact within the digital economy.
Platforms are not a new business model, but rather an old one that has been rejuvenated
by the sheer scale and scope of the participants in digital economy. The complexity this
creates has renewed the need for, and the value in having a simple meeting place where
those interested in trading particular products and services can find one another, and
perhaps be entertained while doing so. It appears that users are not looking for a
particular seller, or someone that carefully selects and assures the quality of suppliers,
instead they crowdsource recommendations and ask only that they be able to search for,
or introduced by algorithm to, the best possible match.
Many digital marketplaces remain free to consumers, the market-makers having decided
against charging for entrance or use of their platform services, and instead to use the
available technology to monetise the information conveyed by users. While this was not
possible in the past, it is now, largely as a result of the ability to digitalise what we know
(the customer relationship), and the low value that users attach to the sharing of this
information. This does not mean competition is necessarily working effectively, however
nor does it mean that there is undetected anticompetitive conduct by firms. More likely,
the answer lies in consumers having greater awareness of the surplus that is generated,
and more effective tools to extract it from the market when prices hit zero.
To investigate whether the antitrust toolkit remains fit-for-purpose the OECD Competition
Committee held a Hearing in June 2017. This asked whether the tools traditionally used to
define markets, to assess market power and efficiencies, and to assess the effects of
exclusionary conduct and vertical restraints, remain sufficient to address those questions in
the context of these multi-sided platform markets. At the hearing a range of expert
economists from agencies, academia, and private practice were invited to make practical
methodological proposals on how these tools might need to be re-designed or re-interpreted
in order to equip competition agencies with the analytical tools they require when analysing
multi-sided platform markets. This report features each of the contributions made by those
experts (and their co-authors) along with an opening synthesis chapter by the OECD.
What we heard at the hearing was that platforms were different in nature from traditional
markets, and particularly that there were important demand externalities from one side of
the platform to the other (‘cross-platform network effects’) which if ignored could lead to
bad decision-making. However, we also heard that where these externalities were
recognised, the existing tools could be adjusted to account for them. Therefore, where
there is a plausible cross-platform network externality, the most important takeaway is for
competition agencies to consider the value of adopting a multi-sided approach, and to
explain the rationale when deciding not to do so.
In addition to this, there were a number of key messages. The first key message was that
market definition is a less valuable tool in these markets. Nevertheless, where it is a
requirement, rather than an analytical tool, the most effective framework remains the
hypothetical monopolist test, even in the presence of zero prices. On market power, we
heard two key messages. Firstly, that the more sophisticated tools need to be adjusted to
estimate the impact that a price rise on side A of the platform would have on: the demand
from users on side A; the demand from users on side B; and the price that is set on side B.
So for example, surveys and demand estimations need to estimate those elasticities.
Secondly, less sophisticated tools for measuring the market power of a platform also need
adjusting to reflect the existence of a second or third side. For example, shares of volume
on one side can only be interpreted in parallel with shares on the other side, and
profitability must be taken at a platform level and not on sales to just one side of the
market.
The key message on exclusionary conduct was that it should not be assumed to be
harmless simply on the basis that there was another side to the market. If anything
platform markets may provide particularly fertile ground for exclusionary behaviour and
so merit greater scrutiny. A second key message was that while the framework for
assessing the exclusionary effects of exclusivity clauses remains robust, price-cost tests as
a whole are not fit-for-purpose as a tool for identifying predatory pricing in these markets.
A proposed replacement was to consider whether the price would have made sense if it
did not weaken its rival. This might be tested by estimating elasticities and then removing
any substitution effects from the platform’s optimal price setting problem.
Finally, the analytical framework and tools used to analyse efficiencies and the effects of
vertical restraints on a case-by-case basis each remain effective. Indeed, there would
appear to be significant scope for efficiencies to arise in platform mergers to the extent
that they are necessary to combine separate user bases and increase interoperability.
Similarly, where cross-platform network effects are strong there may be a real risk that if
they have the opportunity, users on either side might free-ride and bypass the platform.
As a result there may be significant scope for vertical restraints imposed by the platform
to generate efficiencies by protecting its viability.
Frédéric Jenny
Chairman, OECD Competition Committee
Table of contents
Foreword ................................................................................................................................................ 3
Part I. Introduction and key findings .................................................................................................. 9
1. What are multi-sided markets? Why are they different? ................................................................. 9
2. Market definition ........................................................................................................................... 12
3. Market power ................................................................................................................................. 16
4. Exclusionary conduct ..................................................................................................................... 21
5. Efficiencies .................................................................................................................................... 24
6. Vertical restraints ........................................................................................................................... 26
Notes .................................................................................................................................................. 29
References.......................................................................................................................................... 33
Part II. Market definition ................................................................................................................... 35
1. Market definition in multi-sided markets.................................................................................. 37
1. A working definition of a two-sided market .................................................................................. 37
2. A useful distinction among two-sided markets .............................................................................. 38
3. Assessing the two-sided nature of the market................................................................................ 39
4. Defining one or two markets.......................................................................................................... 41
5. Considering both sides of the market............................................................................................. 43
6. The SSNIP test and the HM test .................................................................................................... 45
7. Conclusions.................................................................................................................................... 49
Notes .................................................................................................................................................. 49
References.......................................................................................................................................... 53
2. Market definition in multi-sided markets.................................................................................. 55
1. Introduction.................................................................................................................................... 55
2. One single market vs. separate markets for distinct market sides.................................................. 56
3. Product market definition with multi-homing and single-homing ................................................. 60
4. Further challenges when applying traditional methods for market definition
in multi-sided markets ....................................................................................................................... 62
5. Conclusion ..................................................................................................................................... 64
Notes .................................................................................................................................................. 64
Part III. Market power ....................................................................................................................... 69
3. Measuring market power in multi-sided markets .................................................................... 71
1. Introduction.................................................................................................................................... 71
2. Features of multi-sided markets ..................................................................................................... 71
3. Practical steps when considering measuring market power in multi-sided markets ...................... 73
4. Measures of market power ............................................................................................................. 74
5. Assessing the strength and impact of indirect network externalities and feedback loops ............. 78
6. Conclusion ..................................................................................................................................... 80
Annex. Examples of cases assessing market power in multi-sided markets...................................... 81
Notes .................................................................................................................................................. 83
By Chris Pike *
Since the turn of the century, economists have understood that multi-sided markets
function in ways that are importantly different from standard markets. Since the ground-
breaking work on the topic by Rochet & Tirole, huge progress has been made in
modelling these markets and the way they work, and identifying the mistakes that can be
made by treating them as traditional markets.1 Naturally, this has consequences for the
way in which competition agencies analyse these markets, and hence on whether, and if
so how, they decide to intervene in these markets. The speed and extent of growth in the
digital economy in over this same period has made this one of the most important,
pressing and analytical challenges that competition agencies now face. This is because
much of that digital growth has been driven by the appearance and expansion of
globalised platforms that disintermediate standard markets and directly connect users,
transforming them into more complex multi-sided markets.
In June 2017, the OECD Competition Committee held a Hearing that looked at whether
the tools traditionally used to define markets, to assess market power and efficiencies, and
to assess the effects of exclusionary conduct and vertical restraints, remain sufficient to
address those questions in the context of multi-sided markets. It then invited practical
methodological proposals from a range of expert economists from agencies, academia,
and private practice on how these tools might need to be re-designed or re-interpreted in
order to equip competition agencies with the analytical tools they require when analysing
multi-sided markets.
While economists typically referred to “two-sided markets” to begin with, we here follow
the recent trend by referring here to “multi-sided platforms”. 2 We do so for two reasons.
Firstly, it helpfully distinguishes between the product of the firm (the platform), and the
*
This paper was prepared by Chris Pike, Competition expert at the OECD Competition Division,
with invaluable comments from Antonio Capobianco, Pedro Gonzaga and Antonio Gomes. The
opinions expressed and arguments employed herein do not necessarily reflect the official views of
the Organisation or of the governments of its member countries. All documents related to this
hearing can be found at www.oecd.org/daf/competition/rethinking-antitrust-enforcement-tools-in-
multi-sided-markets.htm. The experts at the hearing were: Lapo Filistrucchi, Arno Rasek (with co-
author Sebastian Wismer), Kurt Brekke, Kate Collyer (with co-authors Hugh Mullan and Natalie
Timan), Michael Katz, Tommaso Valletti (with co-authors Andrea Amelio and Liliane Karlinger);
Jorge Padilla (with co-author Enrique Andreu), Howard Shelanski (Samantha Knox and Arif Dhilla),
Paul Johnson, and Cristina Caffarra (with co-author Kai-Uwe Kühn). Except where indicated, the
conclusions reached in this paper do not necessarily reflect the views of these experts. The experts
were provided with an opportunity to clarify any views that are attributed to them.
relevant market, or markets, in which the platform operates. Secondly, it accounts for the
fact that while the multi-dimensionality begins with two-sidedness (in which consumers
and sellers meet on a platform), this is only the beginning, and many of these markets
have three sides (consumers, content suppliers, and advertisers) and some even have four
(for example in payment cards) or more.
Examples of multi-sided platforms abound: TV and newspapers that connect viewers and
advertisers; payment cards that connect card holders, merchants, card-issuing banks and
acquiring banks; stock exchanges that connect buyers and sellers; shopping centres that
connect retailers with shoppers; digital platforms that connect users, content providers
and advertisers; listings magazines/directories that connect businesses and customers;
estate agents that connect house sellers and house buyers; and telecom networks that
connect fixed and mobile phone users. They might also be thought to include hospitals
that connect physician groups with health insurers (and even health insurers that connect
hospitals and patients), banks that connect depositors and savers, and supermarkets that
connect producers and shoppers.
There are various definitions of the multi-sided markets in which multi-sided platforms
compete, however, most share the same basic elements, and can be captured as follows: a
market in which a firm acts as a platform and sells different products to different groups
of consumers, while recognising that the demand from one group of customer depends on
the demand from the other group(s). 3 Crucially, if this cross-platform network externality is
present,4 this implies that the structure of prices that the platform sets will determine
volume, not just the level at which it sets the price across the different sides of the market. 5
While the existence of a cross-platform network externality is binary, there was common
agreement amongst experts at the Hearing that there is little value in using this as the
distinguishing feature of a multi-sided platform for antitrust purposes. This is because it is
the magnitude of the cross-platform network externality that determines how big a
mistake it is to overlook it and treat the product as one-sided. Therefore, while a wider set
of markets may exhibit small cross-platform network externalities, the externalities will
only be large enough to be important for the analysis in a smaller set of markets.
Using a bright line to identify when to use a multi-sided approach therefore risks
overcomplicating the assessment of what are, in effect, one-sided markets. However, the
alternative conclusion that ‘multi-sidedness matters when it matters’ means that the multi-
sidedness of a market may depend on the nature of the investigation. For example, the
platform nature of a supermarket may not matter in the context of a local supermarket
merger where the impact on suppliers might be minimal given the level at which supplier
decisions are taken and simple quality measures such as the range of products that are
offered to consumers may suffice. However, if the investigation is into the
anticompetitive nature of ‘slotting fees’ charged by supermarkets to suppliers for greater
prominence on its shelves, then a multi-sided perspective might help explain the rationale
for the practice and hence be invaluable to the analysis. Therefore, where there is a cross-
platform network externality, the value of adopting a multi-sided approach should at least
be considered, and the rationale for deciding not to do so explained.
There are also some important differences between different types of multi-sided
platforms. The first is between those platforms that can observe when a transaction is
taking place on the platform and those that cannot. Where the platform can observe a
transaction, it may charge a price for it if the externality derives from additional use of the
platform by other sides, rather than solely from additional membership. This might be
instead of, or in addition to, any subscription fee that it sets for members.
Within the category of non-transaction platforms, we can think of there being non-
transaction matching platforms, and non-transaction audience-providing platforms. For
example, where the cross-platform network externality is positive on both sides and the
objective of the platform and all users is to find the best possible match, Rasek & Wismer
describe a platform as a matching platform (Shelanski, Knox & Dhilla refer to these as
service-based platforms). A matching platform can be a transaction matching platform if
the transaction is observable (e.g. uber, stock exchanges), but if it is not observable then it
can be considered a non-transaction matching platform (e.g. dating apps, real estate
platforms, Wikipedia).
If the externality runs in just one direction, Rasek & Wismer consider the platform an
audience-providing platform (Shelanski, Knox & Dhilla identify these as subsidy-based
platforms). We can think of these audience-providing platforms as being either
transaction or non-transaction platforms depending on whether the transaction is
observable or not. Typically, an advertising platform (e.g. newspapers) will not be able to
observe the transaction (whether the advert resulted in a sale to a specific customer).
However this is already changing in online advertising where a purchase can be traced
using the trail that is created when a consumer clicks through from an advert and makes a
purchase. In that case, the effect of the advert may become observable to the platform,
which in turn allows it to charge for a commission on the follow-on transaction.
While a two-sided market can be categorised using these distinctions, as Shelanski, Knox
& Dhilla point out, many digital platforms are three-sided and so can be characterised
both as matching two sides that each generate positive externalities (users and content
providers), whilst also providing an audience for a third side that might not deliver
positive externalities (advertisers). The transactions between these three sides may all be
observable or none of them might be.
The nature and strength of the cross-platform network effects is therefore more important
to the analysis than the category of platform. For example, the consequences of some
platforms’ actions can be much greater than they appear at first sight. For example, when
a strong cross-platform network externality exists on more than one side of the market,
this creates feedback loops. In these loops, an action can trigger a spiral of reactions,
which, as in a multiplier effect, increase the magnitude of the consequences of the action.
As an example, increasing the price that users pay might reduce the number of users, but
this may also reduce the value of the platform to advertisers and hence reduce the amount
that advertisers are willing to pay. In turn, this may reduce the return that content
providers earn when their content is viewed on the platform, thereby reducing the amount
or quality of content, which may reduce the number of users. Once again, this may then
reduce the amount that advertisers are willing to pay, and so forth. Each action the
platform takes can therefore create a series of reactions (a ripple effect). If these effects
go far enough they may tip the firm towards failure on the one hand, or dominance
(monopoly) on the other.
2. Market Definition
A traditional starting point for framing an analysis of the competitive effects of a merger,
an action or an agreement is to define the relevant market(s) that might be affected. This
can help to identify demand and a set of relevant competitors. However, when a merger,
action or agreement involves either a multi-sided platform, or a firm that trades with a
multi-sided platform, there is a preliminary question of how many markets to define. For
multi-product or multi-location firms, the answer is the result of the market definition
exercise, which identifies the scope of the market, and hence whether those different
products and locations fall within the same or different markets. In contrast, for multi-
sided platforms, the product that a platform provides to one side of the market does not
compete with the product it provides to another side. In the case of multi-sided markets
the question of how many markets to define cannot be answered within a market
definition exercise, instead it is a conceptual question that requires an answer before any
exercise to define the scope of the market can be carried out.
excluded. Notably this would mean that non-transaction platforms would be defined as
competing in a single two-sided market rather than two interrelated markets.
increase one and leave one unchanged (or vice-versa), it might increase one and reduce
the other (or vice-versa), or it might increase both.
The need to re-optimise means firstly that each iteration of the test on a candidate market
needs to be repeated for each of the ways in which the price(s) might be raised to increase
profitability. Furthermore, the optimal balance of prices might change as the scope of the
candidate market is expanded so the same three options might need to be tested at each
iteration.14 This introduces considerable additional complexity, and, if not tackled, would
lead, as Filistrucchi explains, to a bias that overestimates the size of the market that is
defined, thereby potentially underestimating the market shares of firms within that market.
It may therefore be the case that the complexities of applying the hypothetical monopolist
test are insurmountable, while the alternatives are undesirable. The first best solution in
such cases would be to leave the market undefined where possible. However, if defining a
market is unavoidable, and as is often the case, the SSNIP/SSNDQ test cannot be
operationalised, the best option is to use the hypothetical monopolist test as a framework
(or thought experiment) onto which qualitative evidence is applied (for example views on
substitutability from consumer groups, industry analysts or firms that are informed by
verified observations on previous experience). This prevents the exercise from slipping
into a characteristics-based process, which takes no account of substitutability.
There might be little value in carrying out a market definition exercise in markets
involving multi-sided platforms. Therefore, consider carefully whether a market
definition exercise is a necessary and proportionate use of resources.
When defining markets is an unavoidable requirement, first decide how many
markets to define;
• An assessment of the significance of the cross-platform network effect should
be used to identify those markets that should not be treated as traditional one-
sided markets.
• For the purposes of a competitive assessment there is little meaningful
distinction between defining a two-sided market and defining two interrelated
markets, as long as the effect of the cross-platform network effect is
recognised and analysed. However, in some jurisdictions the choice may
have an important effect on which efficiencies the legal analysis allows to be
weighed against any loss of competition that is identified. Therefore, where
cross-platform network effects are important, and a market definition is
required, defining a single two-sided market ensures that the assessment as a
whole is based on the full set of possible competitive and efficiency effects,
and no effect is arbitrarily excluded. Notably this means that non-transaction
platforms would be defined as competing in a single two-sided market rather
than two interrelated markets.
When defining the scope of the market(s);
• The framework of the hypothetical monopolist test provides a discipline that
helps guard against the adoption of a characteristics-based approach to
market definition.
3. Market power
However, there are also challenges to using surveys, since identifying particular effects
while holding everything else constant may not be straightforward. For example, we
would need to assess the three effects set out by Brekke above. To assess the first effect,
Filistrucchi suggests that sellers (e.g. hotels) might be able to tell us how a change in
commission would affect their demand for the platform. However, consumers are
unlikely to be able to tell us how a change in the commission that sellers pay the platform
would affect their demand for the platform. In order to assess this second effect, we might
therefore need to ask how consumers’ demand for the platform would react to the change
in the number of sellers on the platform (or any change in sellers’ prices that is passed
through) when the commission increases. Fortunately, we should know this change from
the sellers’ response that we obtained when quantifying the first effect.
To estimate the third term (the rebalancing effect), a survey would also need to ask the
platform how it would change the price it charges consumers (or the quality it sets), if its
commission on sellers were to increase. However, there might be a question mark over
the platform’s incentive to provide a genuine estimate of this figure. It might therefore be
necessary to validate the figure without input from the platform itself. This might be
possible, but would not be straightforward. We would need, for example, to know the
change in the quantity of sellers (or sales) that would maximise profits for the platform if
it were charging a higher commission. We could then identify the change in consumer
demand that would trigger that size of increase in the quantity of sellers. Finally, we
would need to know how much lower the price to consumers would need to be to trigger
the increase in consumer demand that would set this chain in action.
Where these methods are effective and elasticities are successfully estimated, these
estimates can be plugged into the reformulated UPP and GUPPI expressions that Brekke
identifies. However, in a non-merger context in which the authority wants to understand
the level rather than the change in market power, they can also be plugged into an
adjusted Lerner index to provide a measure of a platform’s market power. Where these
estimates are not available, a potential short-cut set out by Tremblay (2017) is to compute
this adjusted Lerner index using administrative data on profits, fixed costs and revenues.
Where this administrative data is available, an adjusted Lerner index can be calculated as:
the total profit of the platform, plus the fixed costs of the platform, all divided by the total
revenue of the platform.
Other tools
Market shares, barriers to entry and exit, measures of concentration or profitability, and
patterns of use (e.g. single or multi-homing) are each also used to help assess market
power. However, the traditional problems of these types of tools that are not based on
consumer responsiveness, are exacerbated in a multi-sided context. Firstly, as Brekke
explains, some of these tools may assume no product differentiation, while platforms are
highly differentiated (e.g. strengths in different geographic areas, or amongst different
types of user), and the network effects themselves drive much of this differentiation.
Secondly, as Collyer, Mullan & Timan identify, a meaningful unit of measurement is not
always straightforward; for example, value, capacity, volume, or volume of full priced
sales might each make sense in different circumstances. In non-transaction multi-sided
platforms this can be further complicated if there is no common unit that can be used
across both sides, since this makes it unclear how to synthesize the two. Thirdly, these
tools provide no information on substitutability, and so give no sense of how
(in)vulnerable a given market share is. This is particularly problematic in multi-sided
markets since the cross-platform network effects also provide scope for the observed
market shares to quickly and radically change (tip).
As Collyer, Mullan & Timan suggest, market share tools are therefore of most value
when looked at over a period of time, since this indicates a degree of durability. They can
be of particular value when observed over a period of time during which there was a
change in the relative value of the products (e.g. a price increase). Effectively this
introduces consumer responsiveness into the tool. Of course, where such observations can
be identified in data, they can be turned into event studies, a more sophisticated tool that
can provide insight in a multi-sided context, provided the necessary adjustments are
made. For example, event studies of two-sided platforms need to consider what is
happening on the other side of the market, since the consequences of a reduction in the
value offered by the platform might be clear on one side but not the other.
Another tool that does not require information on responsiveness is to proceed directly to
measure the platform’s profitability and to compare that to a counterfactual of what a
competitive return would be. Collyer, Mullan & Timan point out that in a multi-sided
context this would need to recognise that costs incurred, and profits/losses on the other
side of the market, are part of the profitability of the platform, and need to be assessed
together. Many of the challenges faced in one-sided markets re-surface here. For
example, the difficulty in accurately measuring economic profit as opposed to accounting
profit, and the identification of the relevant counterfactual.
With regard to single-homing or multi-homing, both Collyer, Mullan & Timan and Rasek
& Wismer suggest that it can be useful for agencies to examine patterns of use and
establish whether users on one side of the market tend to single-home or multi-home on
different platforms. This can be important for understanding the nature of competition in
the market, for example, whether firms compete to sell each unit, or instead compete for
exclusive relationships with customers. However, as Rasek & Wismer note, it is not clear
whether the predominance of single or multi-homing suggests in and of itself that the
platform has market power. Widespread single-homing or exclusive use might, for
example, be taken to suggest that consumers do not see other platforms as good
substitutes (if we were to assume that consumers would sometimes use these other
platforms if they considered them a good alternative). However, it does not actually tell
us anything about consumers’ views on the potential substitutability of the platform; in
particular, it might be expensive to multi-home and there might be fierce competition
amongst platforms to be the exclusive platform used by each consumer, or at least by the
marginal consumers. 21
There is also an ambiguity to multi-homing (non-exclusive use of a platform). This might
be interpreted as evidence of users switching their demand between platforms (e.g. using
different supermarkets, search platforms, dating applications or advertising routes),
thereby implying strong substitutability and close competition. However, it might also be
interpreted as evidence that the platforms are complementary, thereby implying little
competition (e.g. using two search engines but using them to search for different things,
or using different advertising routes to reach different single-homing groups of users). 22
It is also possible to take a narrower definition of multi-homing as the use of multiple
platforms when making a single decision. For example, the use of a single platform when
looking to order a takeaway pizza on a Saturday evening might be defined as single-
homing, despite the fact that the consumer uses multiple platforms for food delivery over
the course of a month. Adopting this narrower definition makes multi-homing a closer
approximation of substitutability since it eliminates the possibility that the different
platforms were being used when making slightly different types of decision. However,
information on when multiple platforms are used within the same decision is often more
difficult to obtain. 23 Furthermore, even if multi-homing is common on one-side, it might
not indicate that the multi-sided market itself is highly competitive. For example, it is
often noted that surplus built up from multi-homing users (e.g. advertisers or callers to
mobile phones) can then be competed away on attracting single-homing users (e.g.
readers or mobile phone contract holders). However, if there are constraints that prevent
the platform offering negative prices to single-homing users, then the platform might be
able to limit the extent to which it competes away the surplus that it extracts on the other
side of the market. 24
Where tools are not based on consumer responsiveness, care is needed in interpreting
what an observed pattern of use says about substitutability on that side of the market, and
more generally what substitutability on one side of the market implies for the platform’s
market power, which needs to be judged across all sides of the market. 25 Nevertheless,
these tools might, as Collyer, Mullan & Timan suggest, be used to conduct a preliminary
analysis that considers the difficulties that arise as a result of the multi-sided nature of the
market (see above), and which is then adjusted in a second stage of the assessment to
reflect the impact of any cross-platform network effects. Where the cross-platform
network effects are one-way, the preliminary analysis can be sufficient to conclude on the
degree of market-power held by the platform in the provision of a product that generates
no cross-platform network effects for the other side of the market. However, where
products generate two-way cross-platform network effects, the preliminary view on the
market power of the platform will need to be revised. This revision requires an
assessment of whether the cross-platform network effects increase or decrease the degree
of market power identified in the preliminary assessment, and by how much. 26
Where strong cross-platform network effects run in both directions, it is not possible for a
multi-sided platform to have market power on one side of the market. Either it has a
degree of market power as a platform, or it does not. Substitutability of demand might be
different on either side, but given the interrelationship of pricing across the platform, it is
not meaningful to conclude that a platform has market power on one-side of the platform.
For those tools that measure market power based on the responsiveness of demand, cross-
platform network effects need to be integrated within the analysis from the start.
• There are at least six effects that need to be estimated in order to apply the UPP
indices (or GUPPI) that have been adjusted for use in multi-sided markets. These
include the full impact that a price rise on side A will have: 1) the effect on demand
from users on side A; 2) the effect on demand from users on side B; and, 3) the
effect on the price on side B. They also include the same three impacts that a price
rise on side B would have. These six effects can be estimated by surveying users on
each side of the platform, though the questions will need testing with the relevant
audience.
• Where data and time permits, estimates of these effects can also be obtained from
demand estimations that can be used to simulate the effects of a merger or to
estimate an adjusted Lerner index.
• In cases where estimates of diversion ratios and elasticities are unavailable, it may
be that there is adequate administrative data to compute the adjusted Lerner index
using data on profits, fixed costs and revenues. Where this data is available, a
generalised Lerner index can be calculated as: the total profit of the platform, plus
the fixed costs of the platform, all divided by the total revenue of the platform.
• Since quantifying cross-platform network effects is a key task for the assessment of
both competitive effects and efficiency effects in multi-sided platform cases, it may
be worth collapsing these two stages into a single exercise in which both the agency
and the firm(s) seek to quantify the cross-platform network effects.
For other tools that measure market power without reference to the responsiveness of
demand, for example those that measure concentration or profitability, the impact of
cross-platform network effects might be reflected in a second stage of the assessment,
after a preliminary analysis has been conducted.
• The preliminary analysis might use standard tools to identify: the percentage of
users that use the platform; barriers to entry and exit; and profits. It might also look
at the patterns of single and multi-homing behaviour by users since these can be
helpful for understanding the nature of competition in the market. Taken together,
these analyses might allow a preliminary view on the market power of the platform.
• However, care is needed in interpreting what an observed pattern of use (e.g. single-
homing) says about substitutability on that side of the market, and more generally
what substitutability on one side of the market implies for the platform’s market
power, which needs to be judged across all sides of the market.
• Where cross-platform network effects are one-way, this preliminary analysis can be
sufficient to conclude on the degree of market-power held by the platform in the
provision of a product that generates no cross-platform network effects for the other
side of the market.
• Where products generate two-way cross-platform network effects, the preliminary
view on the market power of the platform then needs adjusting to reflect these cross-
platform network effects. This requires an assessment of whether these effects
increase or decrease the degree of market power identified in the preliminary
assessment, and by how much.
4. Exclusionary conduct
It might be argued that multi-sided markets require less scrutiny from antitrust authorities
and should be treated more leniently. This is because cross-platform network effects
magnify competitive constraints suggesting that these platforms have less market power
than first appears and because there are clear pro-competitive rationales for building
volume at the expense of rivals to take advantage of network effects.
However, both Katz and Valletti, Amelio & Karlinger emphatically disagree that greater
leniency is required. Katz concludes that the markets in which multi-sided platforms
operate may provide particularly fertile ground for exclusionary conduct, while Valletti,
Amelio & Karlinger suggest that exclusionary practices are more likely in these markets,
rather than less likely. 27 In each case, the conclusion is that examination of exclusionary
unilateral conduct in multi-sided markets should be a greater priority for agencies than it
is in traditional markets.
To assess the effects of exclusivity clauses involves following a framework of inquiry that
explores the impact of the clauses on rivals’ costs, and then on the intensity of
competition. 31 This broad framework remains applicable for cases involving multi-sided
platforms. In contrast, the more specific price-cost tests and recoupment tests often used
in predatory pricing cases no longer appear reliable. 32 A point that was made early in the
development of the multi-sided platform literature was that below cost pricing on one side
is more likely to be pro-competitive in a multi-sided market since it may help the
platform internalise cross platform network externalities. However, both Katz and
Valletti, Amelio & Karlinger here make the distinct point that not only can a platform
predate by reducing its total price to unsustainable levels, but that it can also do so by
changing the balance of prices across the different sides of the market. The implication is
that even adjusting price-cost tests to focus on net price is insufficient. 33 Instead, these
tests remain potentially misleading in multi-sided markets and should not be relied upon.
Katz also argues that the recoupment test needs to be interpreted with care. For example,
he urges agencies not to interpret this as a test of the rationality of below-cost pricing.
Instead, he argues that agencies should ask firstly whether below-cost pricing is profitable
because it makes the platform a stronger competitor by building up its base; and secondly
whether below-cost pricing is profitable because it weakens competition by preventing
rivals building their own user bases. This requires an understanding of whether the
below-cost pricing would have been profitable in a counterfactual world in which that
pricing did not weaken its rivals (for example by reducing its volume), allowing them to
continue to offer the same value product that they would have offered absent the below
cost pricing.
This ‘no economic sense’ test would identify clearly those exclusionary cases where
allegedly exclusionary conduct is harmful in multi-sided markets (while leaving a grey
area for those cases where there is an efficiency rationale but also an anti-competitive
effect). Unlike the as-efficient competitor test, this has the distinct advantage of
protecting consumers when a more efficient platform engages in conduct that excludes a
less efficient platform and reduces competition. As Katz says, there are cases where
competition between an incumbent and a less efficient rival is better for consumers than
facing a monopolist (even one with low costs), and this is true in both one-sided and
multi-sided markets. As such, requiring an investigating competition agency to show that
a firm’s conduct fails the as-efficient-competitor test is inconsistent with an effects-based
approach.
An additional proposal made by Katz is that the tools used to test for recoupment should
consider not only future recoupment opportunities, but the prospects of simultaneous
recoupment, for example on the other side of the market, or in an aftermarket.
5. Efficiencies
As with competitive effects, there is a risk that efficiencies generated on another side of
the market will be missed if the multi-sided nature of the platform is not recognised.
Alternatively, such efficiencies might be identified but ruled to be out-of-market
efficiencies and hence not relevant for the legal assessment. However, as touched upon in
the market definition discussion, efficiencies or anticompetitive effects on other sides of
the market will be relevant whenever cross-platform network effects are significant.
incompatible, it is perhaps surprising that there are no cases in which efficiencies have
been accepted. One answer might be that while efficiencies are more likely to be
generated in multi-sided markets, there often may remain less anti-competitive ways of
achieving the same efficiencies, for example by allowing interoperability or adopting
shared standards. In any case, as Johnson suggests, it would appear that agencies should
give particularly careful consideration to the scope for efficiency defences in multi-sided
markets.
any efficiencies that accrue to advertisers are unlikely to benefit users. This means any
harm to users is unlikely to be counterbalanced by efficiencies to advertisers. Agencies
may therefore focus on evaluating the existence of efficiencies for advertisers in such
cases – for example on efficiencies to users when users are harmed, and on efficiencies
for advertisers when advertisers are harmed.
Where cross-platform network effects are strong, mergers of multi-sided platforms might
be expected to generate efficiencies if they combine separate user bases and increase
interoperability. There would therefore appear to be significant scope for efficiencies to
arise in platform mergers.
Agencies should give careful consideration to the scope for efficiency defences in multi-
sided markets. Focusing analysis on the magnitude and merger specificity of such effects,
rather than their existence may therefore provide better analytical value for agencies.
Standard econometric tools such as event studies can sometimes be used to assess the
efficiencies that have previously been generated by greater scale. These do not require
estimates of the cross-platform network effects.
To use simulation tools to understand the likely efficiencies of a merger for users on each
side of the market, agencies will need an estimate of the cross-platform network effects.
Surveys or demand estimations can be used to generate these estimates, as they were in
the competitive effects assessment.
Operationally there may be advantages to running the competitive effects and efficiencies
assessments as a single effects assessment in those cases where the multi-sided nature of
the market is undisputed.
It may also be a useful operational step to prioritise analytical efforts based on the nature
of relationships in multi-sided markets. For example, in an audience providing (or
subsidy-based) platform, agencies might focus on efficiencies to users when they expect
users to be harmed, and on efficiencies for advertisers when they expect advertisers are
harmed. In contrast, in a matching (or service-based) platform, agencies will need to
consider all sides of the market.
6. Vertical restraints
hence competes with the platform. This may create some challenges as to whether a case
involves price fixing amongst rivals or a vertical restraint.
advertising may indeed no longer be viable if sellers are able to offer cheaper prices on
their own website than on a platform website. However it is unclear what value
consumers place on these investments (provided the platform itself remains viable), and
hence how much these investments would be missed if the business case for them no
longer made sense. This is particularly problematic if the restraints are at the same time
likely to soften competitive incentives and lead to higher prices.
Johnson also identifies an interesting efficiency defence that might arise particularly in
multi-sided platforms. He cites a paper by Lee (2013) which identifies the importance of
exclusivity clauses for smaller videogame platforms that were seeking to enter into the
videogame market. These platforms were able to use the restraints to counteract the
strong cross-platform network effects that incumbents enjoyed which might otherwise
have prevented them from entering and competing on the market. 36 While such entrants
would not hold market power at the time they agreed these clauses, they may later grow
into stronger positions. Therefore, the case-specific context in which the agreements
apply will matter and a form-based approach will be an unreliable indicator of the effect
of the restraint on consumers.
such restraints, and that this preserves a competitive constraint on the commission that is
charged by the platform. If this argument is valid, it suggests that APPAs are mutually
beneficial and hence more likely to exist for efficiency reasons. If however, the number
of users is treated as an input into the product, then the analysis would miss the fact that
users are likely to switch away from platforms if sellers choose to delist. The potential
competitive threat posed by the option to delist would then be missed, and the
competitive constraint on the platform’s commission underestimated. As a result, the
conclusions reached on the effect of the vertical restraint might be different (it might be
judged harmful when it is not). This suggests that while it is certainly true that parallels
can helpfully be drawn between analysis in one-sided and multi-sided markets in order to
explain certain theories of harm, the analysis itself requires a recognition and
understanding of the difference that cross-platform network effects make.
Notes
6 The economic distinction between these categories that we have highlighted above may of course
be relevant in a different context.
7 An exception which simplifies some cases is where users on other sides of the platform are
indifferent to the use (or membership of) of the platform by another side. For example advertising
markets might be analysed as one-sided if readers, viewers or listeners are indifferent to the
quantity and nature of advertising on their product. There also remains of course the question of
the scope of that one-sided advertising market: does it include television, radio, newspapers, social
media and so on? Is it for ages 25-35 or 75+? And over which geographic area? However, these
are traditional market definition questions, which can be answered using traditional tools.
8 See for example, Kaplow (2010 and 2013). Rasek & Wismer and others suggest that it remains
useful.
9 Another example is the Vertical Block Exemption Regulation in the EU, where satisfaction of the
30 per cent market share threshold may hinge on whether one single or two interrelated markets
are defined.
10 In a SSNIP test, the profitability of a small but significant non-transitory increase in price is
examined for each candidate market. If a SSNIP would not be profitable then the scope of the
candidate market is expanded, and the test is re-run on this next iteration of the candidate market.
When a SSNIP is profitable the candidate market is identified as the relevant market.
11 A firm can reduce value and capture surplus by either increasing price, or reducing its costs by
investing less in quality.
12 As Rasek & Wismer point out there is likely to be significant heterogeneity in the cross-platform
network externality for different users and consumers. However, unless the platform can price
discriminate it will need to optimise based on the overall elasticity. If price discrimination is
possible, this might indicate the existence of distinct markets (based on the ability to price
discriminate).
13 Equally, an increase in the price of advertising would reduce demand for advertising in the
newspaper, which may lead to fewer adverts. This, in turn, may increase readership, which would
increase demand for advertising. This feedback effect means that the price increase for advertising
is more profitable than would appear if the impact on readers (and how that in turn affects
advertising demand) were ignored.
14 Note this is different from the cellophane fallacy, which is a problem in one-sided markets that
remains in multi-sided markets. This is the possibility that the market price from which the test
begins is in fact already a monopoly price and hence any increase will not be profitable and so
further increases to the price will not identify a profitable SSNIP since each iteration brings the
price further away from its optimal level.
15 Strictly, however it is worth observing that the barrier to entry is not the cross-platform network
effect itself, but rather the inability of users to co-ordinate their response to that effect. This means
for example, that where users have effective co-ordination mechanisms available to them, this
may remove the barrier to entry, even if the cross-platform network effect remains. This makes
collective switching schemes a potential model for improving the way that these markets work for
users.
16 However, the substitutability of demand might still be different on different sides of the market.
17 This remains the case whether a multi-sided market has been defined, or whether two interrelated
markets have been defined. Though of course, strictly, we do not know if these cross-platform
network effects are strong or not until they have been estimated.
18 See paragraphs 23 to 26 of Brekke, and previously in Affeldt et al. (2013) and Cosnita-Langlais et
al. (2018)
19 Brekke refers to this as a “feedback effect” but we refer to it here as a “rebalancing effect” in
order to distinguish it from the simple feedback loops identified in para 9. It reflects the fact that
the increase in price has a rebalancing effect on the prices set across the different sides of the
market (not only a direct effect on demand on each side).
20 Whilst most practical in longer investigations of unilateral conduct or in the context of market
studies, these estimations are now being used in mergers within sectors that provide both data and
a continuous stream of merger inquiries. For example, both the Netherlands and the UK have
constructed demand models for hospital services, which can be applied in the context of
individual mergers.
21 For example, rival mobile phone networks might compete to be contracted by handset owners, and
would then match those handset owners with those that want to call them (from fixed or mobiles
lines). The networks would then set their price for calling the consumer knowing that the caller
had few good alternative options. However, it would be inaccurate to describe this as market
power without reference to the intensity of competition to contract with the handset owner.
Therefore, the habit of single-homing (having one mobile phone rather than two) might not tell us
much about the market power of the mobile phone network. As discussed in OECD (2017) a
competitive market might be followed by an uncompetitive aftermarket if consumers do not
anticipate future costs (e.g. printer cartridges) or do not incur them (e.g. mobile phone termination
charges under a calling party pays system).
22 For example, users might use a second platform in addition to their usual platform. For example
house renting/sales platforms and general search; or different estate agent platforms for searching
in different geographic areas (or at different price levels). Alternatively, sellers might use a second
platform to reach buyers that single home on that platform (this is the competitive bottleneck).
23 This narrower definition of multi-homing as the use of multiple platforms in the course of a single
purchasing decision is for example used in the CMA’s analysis of the Just Eat / HungryHouse
merger.
24 For example, investments might be required to facilitate paying negative prices and contracting
for exclusive use of a platform.
25 For example, if multi-homing on one-side is interpreted as reflecting complementarity and not
substitutability and hence indicates a lack of market power on that particular side, this might
indicate smaller competitive incentives to compete for consumers to ‘sell’ to the other side.
26 For example, a one-sided assessment might suggest that platform X has a large share of
sellers/advertisers but a small share of buyers. The cross-platform network effects might then
reveal that buyers are relatively insensitive to the range of sellers, while sellers care a lot about the
number of buyers on the platform. This might suggest that another platform with a small share of
suppliers or more consumers might be a stronger constraint than first thought. Alternatively, a
one-sided assessment might suggest that platform Z is in a relatively vulnerable position (e.g. low
barriers to entry, low switching costs, and a small share of users). However, the cross-platform
network effects might reveal that users are very sensitive to the participation of certain sellers (e.g.
important brands), and the platform has a strong position in relation to those sellers (e.g. a high
share or exclusive contracts). This might suggest that the platform has more market power than
first thought.
27 He also notes recent work suggesting that in markets with zero-price (which is not uncommon in
platform markets), anti-competitive tying strategies can be substitutes for predatory strategies.
28 If the cross platform network effect is strong enough, then harm to side A would also harm side B
by reducing participation on side A. However, this may not be the case if these effects are weaker
and in any case users on side B might not foresee the third order effects of their actions.
29 Shapiro (1999).
30 For example, going from 91 to 93 percent of web searches might be unlikely to improve a
platform’s algorithms in the same way that going from one to 3 percent might do.
31 See OECD Fidelity Rebates (2016) for details of this framework.
32 See Wright (2004).
33 See Behringer and Filistrucchi (2015).
34 See Valletti, Amelio & Karlinger citing Segal and Winston (2000) on divide and conquer
strategies, Katz citing Calzolari and Denicolò (2015), and Farrell (2016) on vertical collusion.
35 Johnson is agnostic on the issue of whether restraints are more or less likely to be anticompetitive
in multi-sided markets.
36 Lee (2013).
37 Including for example any contracting externality.
References
Affeldt, Filistrucchi, and Klein (2013), “Upward pricing pressure in two-sided markets” The Economic
Journal, 123(572), 505-523
Armstrong (2002) “Competition in two-sided markets” Mimeo, August 2002
Armstrong (2006), “Competition in two-sided markets” RAND Journal of Economics, 37(3), 668-91.
Behringer and Filistrucchi (2015), “Areeda-turner in two-sided markets” Review of Industrial
Organization, 46(3), 287-306.
Calzolari and Denicolò (2015), “Exclusive contracts and market dominance” The American Economic
Review, 105(11), 3321-3351.
Chandra and Collard-Wexler (2009), “Mergers in two-sided markets: an application to the Canadian
newspaper industry” Journal of Economics and Management Strategy, 18(1), 1045- 1070.
Cosnita-Langlais, Johansen and Sorgard (2018), "Upward Price Pressure in Two-Sided Markets:
Incorporating Feedback Effects," EconomiX Working Papers 2018-3, University of Paris Nanterre,
EconomiX.
Evans (2003), “The antitrust economics of multi-sided platform markets” Yale Journal on
Regulation, 20, 325–382.
Evans and Schmalensee (2012), "The Antitrust Analysis of Multi-Sided Platform Businesses" Coase-
Sandor Institute for Law & Economics Working Paper No. 623, 2012.
Farrell (2016), “Fidelity Rebates” OECD Roundtable, June 2016.
Filistrucchi, Geradin, Van Damme and Affeldt (2014). “Market definition in two-sided markets: Theory
and practice” Journal of Competition Law and Economics, 10(2), 293-339.
Filistrucchi, Klein and Michielsen (2012), “Assessing unilateral merger effects in a two-sided market: an
application to the dutch daily newspaper market” Journal of Competition Law and Economics, 8(2),
297-329.
Fletcher (2007), “Predatory pricing in two-sided markets: A brief comment” Competition Policy
International, 3(1), 221–224.
Hermalin and Katz (forthcoming), “What’s So Special about Two-Sided Markets?”In Economic Theory and
Public Policies: Joseph Stiglitz and the Teaching of Economics, New York: Columbia University Press.
Kaplow, (2010), “Why (Ever) Define Markets” Harvard Law Review, 124, 438.
Kaplow (2013), “Market Definition: Impossible and Counterproductive” Antitrust Law Journal , Vol. 79,
No. 1 (2013).
Lee (2013), “Vertical Integration and Exclusivity in Platform and Two-Sided Markets” American
Economic Review, 103(7), 2960–3000.
OECD (2015), “Across Platform Parity Agreements” OECD Hearing, October 2015.
OECD (2016) “Fidelity Rebates” OECD Roundtable, June 2016.
OECD (2017) “Competition issues in Aftermarkets” OECD Roundtable, June 2017.
Rysman (2004), “Competition between networks: A study of the market for yellow pages” The Review
of Economic Studies, 71(2), 483-512.
Rochet and Tirole (2002), “Cooperation among competitors: Some economics of payment card
associations” RAND Journal of Economics, 33(4), 549–570.
Rochet and Tirole (2003), “Platform competition in two-sided markets” Journal of the European
Economic Association, 1(4), 990-1029.
Rochet and Tirole (2006), “Two-sided markets: A progress report” RAND Journal of Economics, 37(3),
645-67.
Segal and Whinston (2000), “Exclusive Contracts and Protection of Investments” The RAND Journal of
Economics, 31(4), 603-633.
Segal and Whinston (2000), "Naked Exclusion: Comment" American Economic Review, 90(1), 296-309.
Song (2015), “Estimating platform market power in two-sided markets with an application to magazine
advertising.” Working Paper.
Shapiro, C. (1999). “Exclusivity in Network Industries” George Mason Law Review, 7(3), 673–83.
Tremblay (2017), “Market Power and Mergers in Multi-Sided Markets”, Working Paper June 2017.
Weyl (2010), “A price theory of multi-sided platforms” American Economic Review, 100, 1642-72.
By Lapo Filistrucchi *
Abstract
Many authors have proposed different definitions of a two-sided market. While the debate
may not be fully settled, for all practical purposes a good working definition1 is that a two-
sided market is a market in which a firm acts as a platform and sells two different products
or services to two groups of consumers, while recognising that the demand from one group
of customers depends on the demand from the other group and, possibly, vice versa. 2
Importantly, the demands on the two sides of the market are linked by indirect network
effects 3 and the firm recognises the existence of (i.e. internalises) these indirect network
effects.
The buyers of the two products, however, do not internalise these effects, which are
therefore often called externalities.
Although firms’ strategies in two-sided markets may be, under some conditions4, similar to
those in one-sided markets with complementary products, the fact that buyers do not
internalise these externalities makes a two-sided platform different from the case of
complementary products5. In the case of complements, both products are bought by the same
* Lapo Filistrucchi, Department of Economics and Management, University of Florence, Italy and
CentER and TILEC, Tilburg University, The Netherlands. Corresponding Address: Lapo
Filistrucchi, Dipartimento di Scienze per l’Economia e l’Impresa (DISEI), Università di Firenze,
Via delle Pandette 9, I - 50127 Firenze, Italy, e-mail: lapo.filistrucchi@unifi.it. This contribution
draws from previous work with my co-authors Pauline Affeldt, Eric van Damme, Damien Geradin
and Tobias Klein. I am indebted to them and, more generally, to TILEC members for help in
developing and clarifying the ideas I here put forward.
buyer, who, in her buying decision, can therefore be expected to take into account both
prices. Customers of a two-sided platform do not typically take into account both prices. 6
Typical examples of two-sided platforms include (i) media companies, that sell content
and advertising space, (ii) payment cards companies, that sell the use of a card to buyers
and that of a point-of-sale (POS) terminal to shops, or (iii) online intermediaries, that sell
their services to buyers and sellers.
In media markets, advertisers’ demand for ads on a media outlet increases with the
number of consumers of content (viewers, readers, listeners, etc.), while the latter might
also be, positively or negatively, affected by the quantity of advertising. Similarly in
payment cards markets, the more cardholders there are, the higher the demand from shops
and vice versa. Card issuers such as American Express or VISA are well aware of this
relationship between the two demands they face. Also online intermediaries such as eBay
know that the more buyers visiting their website, the more likely it is that sellers will use
their services and vice versa. In fact, the most common business model on the Internet, as
shown by the success of Google or Facebook, is to attract users with various free services
and sell their attention to advertisers.
For instance, assuming that a customer holds a card and a shop has the corresponding
point-of-sale terminal, even if this customer wants to pay by card, the merchant has to be
willing to accept that card for that particular transaction and vice versa. Once again these
externalities are not internalised by the users of the platform, i.e. the cardholder and the
merchant. For instance, suppose a given merchant would benefit from being paid by card
because she would not need to go to deposit cash and she would not have to face the risk
of being robbed. A cardholder would not take that into account when offering to buy in
cash or by card. He would only consider his own convenience.
In a two-sided market, where two products or services are sold to two groups of
customers, one can define the two distinct concepts of price level and price structure 11.
The price level is (roughly) the sum of the two prices, while the price structure is
(roughly) the ratio of the two prices.
For a market characterised by a transaction between end-users to be two-sided, it is also
necessary that, not only the price level, but also the price structure affects the volume of
transactions. 12 For that to be the case, it needs to be impossible for the side that pays more
to the platform to pass through the difference in price to the other side. If a complete
pass-through were possible, the price structure chosen by the platform would not matter.
The platform would not control the relative price charged to the two sides.
Clearly, a complete pass-through can only take place if there is a transaction between
customers on both sides of the market. Only in those markets there may be market
conditions such that the market is in fact not two-sided 13.
In markets where there is no transaction between end-users of the platform, no pass-
through between the two sides can take place. Thus, the platform has perfect control of
the relative prices charged to the two sides. 14
Before being concerned with how to perform market definition when the market is two-
sided, we should assess whether the market is in fact two-sided and, if so, whether two-
sidedness is likely to matter.
In order to assess the two-sided nature of the market, it is crucial to identify and characterise
the indirect network effects that link the demands on the two sides of the market.
One might therefore ask whether such indirect network effects exist, whether they are one
or two, whether they are both positive, or one is positive and one negative and, finally,
how significant they are.
For instance, when analysing a merger in the newspapers market, one might want to
know whether a larger readership of a newspaper ceteris paribus (i.e. holding constant
also prices) implies a higher demand to advertise on that newspapers, whether readers
dislike advertising and, if so, whether advertisers like readers more than readers dislike
advertising. Similarly for a merger among TV channels.
If a market is a non-transaction market, looking at externalities is sufficient.
If instead the market is a transaction market, then one should also check if there are
transaction costs or, more generally, limits to the bilateral setting of prices among buyers and
sellers or if there are platform constraints on pricing between customers on the two sides.
In payment cards markets, for instance, this could be the case not only of the no-
surcharge rule but also of menu costs for a shop that wishes to set a different price for its
products depending on whether the buyer pays by cash, by VISA debit, VISA credit or
AMEX. But it could also be the case of a shop that faces a lot of competition from shops
nearby and therefore has a high probability of losing a customer when attempting to
surcharge.
Only if these constraints exist then the market is two-sided, because the side charged the
higher price by the platform would be unable to pass through completely the difference in
prices to the other side.
Indeed, the lower the pass-through among the parties that transact, the more important the
two-sided nature of the market.
In practice, in order to assess the two-sided nature of the market, both qualitative and
quantitative approaches are possible.
As a first step one could use a qualitative approach and focus on checking whether there
are indirect network effects and, if so, what their sign is, i.e. whether these effects are
both positive or one is negative.
For instance, one might want to know not only whether advertisers base their decisions on
which newspaper to place their ads on the number of readers and whether indeed they
attach positive value to a higher readership, but also whether readers like, dislike or are
indifferent to advertising.
If they are not present, one could then proceed considering the market(s) one-sided.
If instead indirect network effects are present, one needs to distinguish :
• If the market is a non-transaction one, since the pass-through between end-users is
by definition zero, the market is two-sided.
• If the market is a transaction one, one should check to what extent transaction
costs, or constraints set by the platform, limit the possibility of pass-through
between the two sides. If there is scope to believe that the pass-through is high,
then one could come to the conclusion, that although the market is two-sided, the
two-sided nature of the market might not play a great role in practice.
The simplest way to assess qualitatively the two-sided nature of a market could in some
cases be a logical argument.
For instance, in the case of newspapers or TV, it would appear evident even at first sight
that advertisers value positively the number of readers of a newspaper or the number of
viewers of a TV channel. Indeed, the only reason advertisers advertise in a newspaper or
on TV is that they aim to reach readers or viewers with their message.
Unfortunately this approach cannot always be followed, as in some cases it is not clear
whether one side cares about the other and a fortiori whether it values the other side
positively or negatively.
For instance, despite some evidence for specific countries, it is not clear what the attitude
of readers is towards advertising in different media.
In fact, one of the drawbacks of this deductive approach is that it may lead to different
conclusions on the existence and, more importantly, the sign of the network effects.
A slightly superior way to assess qualitatively the two-sided nature of a market could be
interviewing agents in the market (i.e. business people but also their customers) or
making them fill-in a questionnaire with the aim of assessing whether they value,
positively or negatively, the presence of more customers on the other side, and in case of
a transaction market, whether there are factors limiting the platform’s ability to control
the price ratio.
For instance, in the case of newspapers, one could ask newspaper readers whether they
like advertising on the newspaper, whether they are annoyed by it or whether they are
indifferent to it.
In some cases such surveys might indeed already exist.
This is the case, for instance, in many countries where communication or social scholars
run surveys with regard to the use and the perception of media.
The main drawback of this interview approach, and of any qualitative approach, is that it
does not allow one to measure the size of the indirect network effects. Yet the latter is
crucial to establish to what extent indirect network effects play a role in market
definition 15.
Hence, as a second step, one might need to assess the two-sided nature of the market by
using a quantitative approach and turn to checking not only whether there are indirect
network effects and whether they are positive or negative but also on measuring their
size.
For instance, in a case involving newspapers, one might want to know how much
advertisers value an additional reader or, in a case involving payment cards, one might
want to check whether merchants care more about one additional cardholder than a
cardholder cares about one additional merchant having a point-of-sale terminal.
In order to answer these questions one can follow two different quantitative approaches:
the stated preference approach (i.e. designing a survey) and the revealed preference
approach (i.e. collecting actual data). Both are often more time consuming than a
qualitative approach as they require the collection and analysis of data. They would thus
seem more applicable in a second phase of an analysis.
In fact, having already identified two-sidedness using a qualitative approach might help in
figuring out which are the relevant questions to formulate and the relevant data to collect.
The main purpose of market definition is to identify the products that exert competitive
pressure on the products sold by a particular firm or firms, be they firms that plan to
merge, a firm suspected of anti-competitive behaviour or firms that might become the
target of a regulatory intervention. Market definition is therefore an attempt to define a
group of products, which are substitutable to such an extent that the firms producing them
can be perceived as competing against each other, thus constraining each other’s ability to
increase prices.
In a two-sided market, a firm sells two distinct products on the two-sides of the market
and the demands for these products are linked by the presence of indirect network effects.
Firms in a two-sided market can be seen as platforms that need “to get both sides on
board” 16 in order to do business.
The question then arises whether there are two (interrelated) markets to be defined or
only one market encompassing the two sides.
For instance, when analysing a merger among TV broadcasters the question is whether
there is a market for TV or there is a market for advertising (on TV) and a market for TV
content. Similarly, in a case involving payment cards, the question is whether there is a
market for payment cards services or a market for payment cards services to cardholders
and a market for payment cards services to merchants.
It turns out that, whether one needs to define one or two markets, depends crucially on the
type of two-sided markets. More precisely,
• In two-sided non-transaction markets, one should define two (interrelated)
markets.
• In two-sided transaction markets, one should define only one market.
In fact, in a two-sided transaction market the product offered is the possibility to transact
through the platform. It takes the form of two distinct products, one for each side of the
transaction, because such possibility needs to be offered to both sides. Yet none of these
two products is sufficient without the other. A customer on one side can consume his
product only if the corresponding customer on the other side consumes his product too. In
other words, the two products need to be consumed in a fixed 1:1 proportion, as perfect
complements, but by two different consumers.
For example, in the purchase of a pair of shoes through a shop, the merchant cannot
receive money through the POS terminal unless the client has a payment card and is
willing to use it; and vice versa.
Importantly, a two-sided transaction market candidate substitute products constraining the
ability of the two-sided transaction platform to raise prices are not only other platforms,
which offer, to both sides, the possibility to transact but also non-intermediated
transactions.
One of the consequences of defining only one market is that a firm would be either on
both sides of the market or on none. Defining instead two interrelated markets would
allow a platform to be on one side of the market but not on the other. Whether one or the
other outcome is right depends on the type of two-sided market under consideration.
A payment card company such as Diners Club is either in the relevant market on both
sides or on none, for the simple reason that either the transaction between the buyer and
the merchant takes place using Diners Club services on both sides, or it does not take
place through Diners Club. The analysis of a merger between two payment-card
platforms should thus consider, for instance, whether cash transactions or PayPal exert
competitive pressure on these payment card companies.
However, in a case involving TV broadcasters, a product might be in the relevant market
on the advertising side but not on the viewers’ side. 17 For instance, suppose that people
do not regard TV and newspapers as substitutes because they read the latter on the metro
going to work and watch TV at home in the evening. Assuming that advertisers are
interested in reaching each person only once during a day, they will tend to regard TV and
newspapers as substitutes. TV would then be in the same relevant market as newspapers on
the advertising side but not on the viewers’ side. The analysis of a case involving TV
broadcasters should then be allowed to conclude that newspapers exert competitive pressure
on TV in the market for advertising but not in the market for content.
Clearly, in two-sided transaction markets end-users on the two sides can be charged both
a fixed fee for joining the platform and/or a per-transaction fee for using the platform.
Conceptually this feature is not present also in single-sided markets where customers are
charged two-part tariffs, as for instance the traditional market for fixed mobile phone
services. Consistently with previous practice in these one-sided markets, one should
define a single market, in which both membership and usage are sold.
The peculiarity of two-sided transaction markets is not the presence of two-part tariffs.
The differences with respect to a single-sided market are the presence of indirect network
effects between the membership markets on the two sides and the fact that the usage
market is a transaction market linking the two-sides. These differences imply that a single
market encompassing membership and usage cannot but comprise both sides of the
market.
In a two-sided non-transaction market instead there is no transaction and, as a result, there
is not such a strong link in the usage market. In these markets the link among the
membership markets is present, because of the indirect network effects, and needs to be
taken into account when defining the relevant market, but it is not so strong that it implies
the necessity of a single market for the purpose of market definition.
Given the necessity to define a single relevant market ecompassing both sides, it is
obvious that one should consider both sides of the market when defining the relevant in
the case of two-sided transaction markets.
For instance, one should look at both buyers and merchants when one defines the market
for (transactions by) payment cards. It may be that ex post, i.e. after the analysis, one
concludes that one side plays a decisive role in the decision. However, a priori it is clear that
both sides need to agree for the transaction to take place through the payment card company.
Also in the case of two-sided non-transaction markets, competition and regulatory
authorities should take into account both sides of the market when defining the relevant
market 18. Indeed, they should consider the role of the indirect network effects and define
two interrelated markets.
For instance, in a merger among newspapers, one should look also at the advertising side
when defining the relevant market for readers and vice versa.
A platform in a two-sided market needs both sides “on board” and therefore competes for
customers on both sides. How much competition a platform faces in getting customers on
one side also depends on its competitive position on the other and vice versa.
It is well known in the economic literature that product differentiation, whether vertical or
horizontal, relaxes price competition in a one-sided market. 19 Similarly, on each side of a
two-sided market, the degree of competition faced by a given platform depends on the
degree of vertical and horizontal product differentiation on that side.
For example, the level of competition faced by a TV station on the advertising side depends
inter alia on the number of its viewers compared to other TV stations. For instance, if a TV
station has many more viewers than its rivals, one can expect a similar price increase on the
advertising side to lead to a smaller loss in advertising than if the TV stations were closer to
each other in terms of number of viewers. One can argue that from the advertisers’ point of
view TV stations are vertically differentiated in the number of viewers.
Moreover, the level of competition faced by a TV station on the advertising side is also
likely to depend ceteris paribus on the demographic composition of its viewers with
respect to that of the viewers of rival TV stations. To the extent that different advertisers
might value some demographic groups of viewers more than others, TV stations can also
be perceived as horizontally differentiated on the advertising side.
Market definition in one-sided markets typically takes product differentiation as given.
However, in a two-sided market both horizontal and vertical product differentiation is
largely determined by pricing decisions.
From the point of view of advertisers, TV stations are likely to be vertically differentiated
(because they have a different number of viewers) and horizontally differentiated (insofar
as they have different types of viewers). Yet both the number and the type of viewers also
depend on the price charged to viewers (whether positive or zero) and, to the extent that
viewers are annoyed by advertising, on the price charged to advertisers, which contributes
to determine the quantity of advertising in the TV station.
Thus, product differentiation on one side not only affects pricing decisions on that side
(as in one-sided markets), but may also depend on pricing decisions on the other side.
Pricing decisions on the two-sides are interrelated.
Hence, the competitive constraints faced by a platform in its pricing strategies can be
assessed only by taking into account both sides when defining the relevant market.
Moreover, neglecting one side of a two-sided market when the product on that side is
priced at zero is conceptually wrong. In fact, firms are competing also on that side.
For instance, one might think that traditional phone directories, that were distributed for
free, competed only on the advertising side. Yet, if a phone directory raised advertising
tariffs and experienced a drop in listings, it would likely suffer not only a direct drop in
profits but also an indirect drop in usage due to people finding less information in the
directory compared to competing directories. Similarly, if the phone directory
experienced a drop in the number of users, possibly because of the appearance of a
competing product of higher quality for users, it is likely that this would lead to a drop in
demand for ad slots from advertisers. The phone directory may then be forced to lower
the price charged to advertisers and/or experience a decrease in the amount of advertising
and in the corresponding revenues.
By failing to consider all sides in the definition of the relevant market one would then
ignore the real competitive pressure faced by the firms under consideration.
It is only in the particular case of a two-sided non-transaction market with only one
externality, that one could safely perform a market definition exercise, on the side of the
market that does not exert any externality, irrespective of the other side.
For example, in a case involving newspapers, if one finds that advertising has no effect on
the readers’ side of the market, one needs to take into account the advertising market
when defining the readers’ market but one can safely define the advertising market
irrespective of the readers’ market. In fact, in that case, whatever the pricing choices of
publishers on the advertising side, they will not affect the readers’ side. Hence, the
platform on the advertising side of the market will not behave differently from a firm in a
single-sided market facing the same advertising demand.
More generally, when defining the relevant market in the case of a multi-sided non-
transaction market, it is only necessary to consider all the other sides towards which the
side under consideration exerts an externality, either directly or indirectly. 20
The most rigorous conceptual tool used to define the relevant market is the so-called
“Small-But-Significant-Non-Transitory Increase-in-Price Test” 21 (in short the SSNIP
test), which defines the market as the smallest set of substitute products 22 such that a
substantial (usually five or ten percent) and non-transitory (often one year) price increase
by a hypothetical monopolist would be profitable.
Starting from a set of candidate products, the SSNIP test is implemented by first
simulating a given price increase above the current level 23 by a hypothetical monopolist
who owns just one product 24 and, as long as that leads to estimated losses in profits,
progressively increasing the number of products owned by the monopolist and simulating
a price increase of all the products the monopolist owns. When the hypothetical
monopolist does not estimate profits to decline following a small but significant increase
in price, the set of products owned by the monopolist in the last simulation constitutes the
relevant market.
The SSNIP test is often performed by Critical Loss Analysis (CLA), for which formulas
are derived under the assumptions of constant marginal costs and either linear or constant
elasticity of demand. 25 Under these assumptions, performing a CLA is exactly identical to
performing the SSNIP test. 26
In any case, the idea behind the SSNIP test (and thus CLA) is that if the small but
significant non-transitory increase in price is unprofitable, then there exists at least one
close-enough substitute to the product whose price is raised. If so, the two products
should be in the same relevant market. And so on and so forth. Thus, both the SSNIP test
and CLA analysis set an implicit benchmark for substitutability between products to be in
the same relevant market.
In addition, the iterative procedure described above is designed to ensure that a relevant
market is defined as the smallest set of substitute products on which a monopolist would
find it profitable to increase prices by a small-but-significant amount; it thus makes sure
that the market is defined in such a way that a monopolist has market power, which is a
basic requirement of economic theory.
If order to preserve the same logic of the one-sided test, the SSNIP test (and CLA
analysis), should be modified differently according to the type of two-sided market:
• In a two-sided non-transaction market, one should check the overall profitability
of a rise in price on each side of the market.
• In a two-sided transaction market, one should instead check the profitability of an
increase in the price level (i.e. the sum of the prices paid for the transaction by the
two sides).
Ideally, in both cases one should allow the hypothetical monopolist to re-optimise the
price structure following the price increase. 27
Furthermore, in a two-sided transaction market, the SSNIP test should take into account
the changes in overall profits (i.e. the sum of the profits on both sides of the market) and
all feedbacks between the two sides of the market when judging the profitability of a
price increase.
Since positive indirect network effects between the different sides of the platform reduce
the profitability of any price increase, the risk of applying a one-sided SSNIP test, which
does not account for these feedback effects, is that in such cases the two markets may be
defined too narrowly.
Consider a two-sided platform with sides A and B linked by positive indirect network
effects. The application of a one-sided SSNIP test on side A would only account for the
direct effect that a price increase will have on the demand and profits of side A. It would
not account for the fact that a reduction of the number of customers on side A is likely to
lead to a reduction of the number of customers on side B so that, if the price on side B is
kept constant, there will be a loss in profits also on side B. It would also not envisage the
fact that the smaller number of customers on side B will in turn reduce the demand of side
A, and so on. Hence, it would also underestimate the loss in profits on side A. The iterative
procedure of the SSNIP test would then stop too early. Similarly for the application of a
one-sided test on side B. On both sides the market would be defined too narrowly.
In other words, in two-sided non-transaction markets with positive network effects, a one-
sided SSNIP test can provide a lower bound to the relevant market.
If instead one network effect were positive and one negative, the implications of applying
a one-sided SSNIP test, which does not account for these feedback effects, is that in such
cases the market may be defined too broadly on the side that exerts a negative externality
and may be defined either too narrowly or too broadly on the side that bears the negative
externality.
Consider a two-sided platform with side A exerting a negative externality on side B and
side B exerting a positive one on side A. The application of a one-sided SSNIP test on
side A would not account for the fact that a reduction of the number of customers on side
A is likely to lead to an increase of the number of customers on side B; so that, if the
price on side B is kept constant, there will also be an increase in profits on side B. It
would also not envisage the fact that the higher number of customers on side B will in
turn increase the demand of side A, and so on; so that, in the end, it would also
overestimate the loss in profits on side A. The iterative procedure of the SSNIP test
would then stop too late on side A. Hence, on that side, the market would be defined too
large. Similarly, the application of a one-sided test on side B would not take into account
the resulting loss in profits on side A and would overestimate the resulting loss in profits
on side B. The iterative procedure of the SSNIP test may then stop too early or too late on
side B. 28 Hence, on this side, the market may be defined too narrowly or too largely.
In other words, in two-sided non-transaction markets with one negative (and one positive)
network effect, a one-sided SSNIP test can provide a upper bound to the relevant market
on the side that exerts the negative externality and enjoys the positive one. It would not
instead be informative on the side of the market that exerts the positive externality and
bears the negative one.
Only in the presence of a single (positive) externality linking the two-sides of the market
could the traditional SSNIP test (and single-sided formulas for CLA) be safely applied in
a two-sided non-transaction market to define the market on the side that does not exert an
externality on the other.
Some authors have proposed that the SSNIP test (and CLA analysis) be performed
without allowing the hypothetical monopolist to re-optimise the price structure. 29
While using the standard single-sided SSNIP test or CLA formulas would lead to a too-
narrow or too-large definition of the relevant market, adopting a two-sided SSNIP test (or
using two-sided CLA formulas) that do not allow the HM to re-optimise the price
structure would lead to a too-large definition of a market. In fact, not allowing the price
structure to be re-optimised would always overestimate the loss in profits due to the
increase in prices, because by definition the optimal adjustment by the hypothetical
monopolist will tend to reduce such a loss.
Hence, both in two-sided transaction and non-transaction markets a two-sided SSNIP test
that does not allow the hypothetical monopolist to re-optimise the price structure can
provide an upper bound to the relevant market.
Finally, it is often the case in two-sided markets that customers on one side of the market
do not pay. Such a situation may arise both in transaction and non-transaction markets,
but it raises different issues in the two types of markets.
In a transaction market, one mainly needs to predict the likely reaction of non-paying
customers to a price increase. This can usually be done by designing an appropriate
survey of existing customers to elicit their willigness to pay. Once this is measured, in a
two-sided transaction market, the SSNIP test can be performed.
When the market is a non-transaction one, a two-sided SSNIP test can be safely
performed on the paying side of the market. However, on the side where the price is zero
it is not possible to perform a SSNIP test. Here the issue is not only that the reaction of
customers to a price increase is not known, but, more fundamentally, that increasing the
price by 5 or 10% has no meaning when the starting price is zero. Any price increase one
would consider would be arbitrary and change the benchmark with respect to the practice in
one-sided markets and the extension just discussed to the paying side of a two-sided market.
However, if the question of interest is whether the free product is in the same relevant
market with a product that is sold at a positive price, one could envisage performing the
SSNIP test starting from this other product and checking whether the test would lead to
adding the product of interest to the relevant market 30.
If instead the question of interest is whether the free product is in the same relevant
market of another free product, then one cannot resort to the SSNIP test.
In fact, it could be argued that in such a case the SSNIP test might even not make much
sense 31. In general, the price is only one dimension of competition among firms.
Conventionally, competition policy has considered it to be the most important dimension
of competition, leaving aside for instance choices on the variety or quality of the
products. The fact that on one side of the market the price is zero most probably indicates
that, on that side of the market, the most important dimension in which firms compete is
not the price. Most likely, competition takes place on quality or variety.
If the relevant competitive dimension is quality, one could envisage an alternative test,
similar to the SSNIP test, where the HM, rather than increasing price, would be
decreasing quality. Such a SSNDQ test has been proposed already for one-sided
markets 32. The starting assumptions are both that a decline in quality leads to a loss in
customers and that an HM would be more likely to find a decline in quality unprofitable
if the product it sells has fewer or less close substitutes. Then the iterative procedure
would be similar to the one of the SSNIP test.
The proposal of a SSNDQ test has not been very successful. In particular, it has been
argued that, since product differentiation is most often multi-dimensional, it is difficult to
establish what is the relevant quality dimension in practice. 33 More fundamentally, if, as it
is the case in one-sided market, customers are paying for the product, it is not certain that,
in the presence of substitute products, an HM would lower the quality of its product less. 34
However, differently from a one-sided market, on the non-paying side of a two-sided
market, given that the price is zero (and assuming it will remain zero), an HM would
most likely lower the quality of its product less in the presence of substitute products,
consistently with the assumption of the SSNDQ test. 35
On the non-paying side of a two-sided market, one can then envisage a SSNDQ test that
is performed by changing the quality and looking at profitability for an HM.
Importantly, as with the extension of the SSNIP test to two-sided markets, and for the
same reasons, such a test should look at overall profitability (i.e. profitability on both
sides) of the quality decrease and should take into account all feedbacks between the two
sides of the market.
While also in a two-sided market it is difficult to establish what is the relevant quality
dimension in practice, there is an obvious dimension that could be taken into account. In
fact, in a two-sided market, one important dimension of quality is, as already argued
above, the size of the network effect, i.e. the number of (some type of) users on the other
side of the market. Thus, identifying the dimension of quality due to the network effect
may be less contentious than in a one-sided market, once the market is established to be
two-sided and the presence of the relevant indirect network effect has been confirmed.
Hence, if the non-paying side bears an externality (whether negative or positive), one can
envisage a SSNDQ test that is performed by changing the quantity on the paying-side of
the market and looking at the profitability of the change for an HM.
Depending on whether the externality is negative or positive, such a SSNDQ test would
ask the HM to raise the network effect or lower it, respectively. For instance, in a case
involving TV stations, assuming it has been found that TV advertising annoys viewers,
one should ask the HM to raise advertising quantity, while in a case involving traditional
phone directories, having assessed that readers are interested in the amount of listings,
one should ask the HM to lower the number of listings.
Notably, the size of the network effect enjoyed or borne by customers on one side also
depends on the price paid by customers on the paying side of the market. Hence, in a two-
sided market in which one side does not pay, the quality on the non-paying side of the
market also depends on the price paid on the paying-side.
A SSNDQ test on the network effect on the non-paying side of the market would thus be
linked, albeit not equivalent, to the SSNIP test on the paying-side of the market.
More precisely, a high substitution towards a competing product on the non-paying side
of the market as indicated by the above SSNDQ test would contribute to a high
substitution on the paying-side of the market as indicated by a SSNIP test on the latter
side, but it would neither be sufficient nor necessary.
To conclude, although the relevant benchmark would clearly change by switching from a
test on price to a test on quality, such a SSNDQ test would allow competition authoritities
to apply the same logic as a SSNIP test.
Since in practice, the SSNIP test is rarely used in its mathematical form and is most often
seen as a conceptual tool to define the relevant market, such a SSNDQ test may be a
reasonable solution to adress the issue of market definition on the non-paying side of the
market when the candidate substitute product on that side of the market are for free.
When instead one or more of the candidate substitute products are paid for, it may be
preferable to perform a two-sided SSNIP test starting from one of these candidate
products, as discussed above, because in such a case it is harder to assume that price is
not the relevant dimension of competition.
7. Conclusions
Notes
observable online (when one clicks on an online ad to open it) and, in such a case, the platform
can charge for it. However, at best only a delayed transaction is present (when someone who saw
an ad buys the advertised product) and this transaction is usually not identifiable (as it is
impossible to say whether someone bought a product because he or she saw an ad), so that the
platform is unable to charge a fee for it. Only recently, using online tracking technologies, it has
become possible to charge advertisers for online transactions between an advertiser and an
internet user that buys a product online after having seen an online advertisement. The ability to
track purchases resulting from an ad are currently limited but such technological developments
may eventually push some media markets to become two-sided transaction markets.
9 Note however that the fact that a two-part tariff can be charged does not necessarily imply that it
will be charged. Indeed both or either of a membership fee and a per-transaction fee can be
charged. In fact, the crucial point is that a per-transaction fee can be charged. For example, for
most payment cards in Europe and the US, cardholders pay at most an annual fee, while
merchants pay a two-part tariff.
10 Other two-sided transaction platforms are virtual marketplaces, auction houses and operating
systems.
11 See Rochet and Tirole (2006).
12 I write “roughly” because prices on the two sides are in different units of measurement. For
instance, in the case of a newspaper, the cover price is per copy of the newspaper, while the
advertising tariff is per page or per column millimetre. Thus the price level is not simply the sum
of the two prices, but rather the sum of the two prices expressed in the same unit of measurement.
Again, in the case of newspapers the price level is the sum of the cover price and the per-copy
advertising revenues. Similarly, the price structure is the ratio of the two.
13 This will be discussed more in detail in the next section.
14 In practice, a two-sided market without a transaction is just an extreme case of a two-sided
market: one where no pass-through is possible. At the other extreme, when the pass-through is
complete, one finds a one-sided market. In the middle lie many different two-sided markets, those
in which some pass-through is possible, although not complete.
15 This will be discussed further in the next sections.
16 Rochet and Tirole (2006).
17 See Evans and Noel (2005).
18 See also Evans and Noel (2008).
19 Two products are said to be vertically differentiated (or differentiated on quality) when, if faced
with the same price, all consumers would buy one of them (the one with the highest quality). Two
products are instead horizontally differentiated (or differentiated on variety) when, even faced
with the same price, some consumers would buy one of them and others would buy the other
(because consumers have different tastes).
20 Indeed, in a multi-sided platform, side A could exert an externality on side B when customers on
side B value more customers on side A, but it could also exert an externality on side B when
customers on side B care about customers on side C and customers on side C care about
customers on side A. Both cases above would lead to equivalent suggestions with respect to
market definition on side A.
21 In the US, the corresponding test is the “hypothetical monopolist test” (HM test). The two tests are
slightly different. See Werden (2003) for a historical account of the ascent of the HM test.
22 For purely expositional reasons, I refer here only to the definition of the relevant product market
and not to the geographic market.
23 In fact, the current level is assumed to be competitive. This is a drawback of the test giving rise to
the so-called “cellophane fallacy” in one-sided markets. In two-sided markets the fallacy may or
may not arise depending on the sign and size of indirect network effects.
24 One of those of the merging parties in a merger case, one of those owned by the potentially
dominant firm in case of abuse of dominance.
25 Critical Loss Analysis works as follows: first, one calculates the so-called “critical loss”, which is
the maximum percentage loss in sales that can be sustained without a given price increase
becoming unprofitable; second, the “actual loss” is defined as the expected percentage loss
following the same price increase. If the actual loss is higher than the critical loss, it would not be
profitable to increase prices. Vice versa, it would be profitable.
26 CLA formulas are different in the EU and in the US, reflecting the difference between the SSNIP
test and the HM test. See Werden (2002a, 2002b).
27 This is proposed also by Emch and Thomson (2006) for two-sided transaction markets. It is
instead proposed by Filistrucchi et al. (2014) for two-sided non-transaction markets.
28 These results are based on a linear specification for the demand function. Linearity, however, is
often assumed in the application of the SSNIP test. As noted above, existing CLA formulas are
based on such an assumption.
29 See Evans and Noel (2008).
30 In fact, the SSNIP test is not a symmetric algorithm. See Werden (2002b). Hence, this could be
considered a second best solution,
31 See also Evans (2011).
32 See Hartmann et al. (1993).
33 See, for instance, OECD (2013).
34 For instance, in a vertical product differentiation like Mussa and Rosen (1978) or Shaked and
Sutton (1982), the lower quality firm finds it more profitable to lower quality exactly because it
has a competitor with a higher quality: by lowering quality, it differentiates more and relaxes
subsequent price competition.
35 Since in this case there is no price competition, by increasing quality, a lower quality firm would
steal customers from the higher quality firm.
References
Damme, E. van, L. Filistrucchi, D. Geradin, S. Keunen, T. Klein, T. Michielsen and J. Wileur, (2010):
“Mergers in Two-Sided Markets – A Report to the NMa”, Netherlands Competition Authority, pp. 1-183.
Emch E and T.S. Thomson (2006). “Market Definition and Market Power in Payment Card Networks,
The Review of Network Economics, 5(1); 45 – 60
Evans, D.S. (2003). “The Antitrust Economics of Multi-Sided Platform Markets”, Yale Journal on
Regulation, 20(2); 325-381.
Evans, D.S. (2009). “Two-Sided Market Definition”, Market Definition in Antitrust: Theory and Case
Studies, ABA Section of Antitrust Law.
Evans, D.S. (2011). “The Antitrust Economics of Free”, CPI Journal, Vol. 7 (1).
Evans, D.S and M.D. Noel (2005). “Defining Antitrust Markets When Firms Operate Two-Sided
Platforms.” Columbia Business Law Review, 667-702
Evans, D.S. and M.D. Noel (2008). “The Analysis of Mergers that involve Multisided Platform
Businesses”, Journal of Competition Law and Economics, 4(3), 663-695.
Filistrucchi, L. (2008). “A SSNIP Test for Two-Sided Markets: The Case of Media”, NET institute
working paper n°08-34.
Filistrucchi L., (2010), “How many markets are two-sided?”, CPI Journal, Volume 7 (2).
Filistrucchi, L., Geradin D. and Eric van Damme (2013). “Identifying Two-Sided Markets”, World
Competition, 36(1); 33-60.
Filistrucchi L., Geradin D., van Damme E., and Affeldt P. (2014), Market Definition in Two-Sided
Markets: Theory and Practice, Journal of Competition Law and Economics, vol. 10 (2), 293-339.
Filistrucchi L. and Tobias J. Klein (March 2013). “Price Competition in Two-Sided Markets with
Heterogeneous Consumers”, mimeo.
Hartman,R., Teece D., Mitchell W. and T. Jorde, (1993), “Assessing Market Power in Regimes of Rapid
Technological Change” Industrial and Corporate Change, vol. 3(2), 317-350.
Mussa, M. and Sherwin Rosen (1978), “Monopoly and Product Quality”, Journal of Economic Theory,
vol.18, 301-3017.
OECD (2013). “The Role and Measurement of Quality in Competition Analysis”, OECD.
Rochet, J-C and J. Tirole (2003). “Platform Competition in Two-Sided Markets”. Journal of the
European Economic Association, 1(4), 990-1029.
Rochet, J-C and J. Tirole (2006). “Two Sided Markets: A Progress Report”. Rand Journal of Economics,
37(3), 645-667.
Shaked, A. and J. Sutton (1982), “Relaxing Price Competition through Product Differentiation”, Review
of Economic Studies, Vol. 49 (1), 3-13.
Werden, G.J. (2002a). “Beyond Critical Loss: Tailoring Applications of the Hypothetical Monopolist
Paradigm”, US DOJ Antitrust Division Economic Analysis Group Discussion Paper No. 02-9.
Werden, G.J. (2002b). “Market Delineation Algorithms Based on the Hypothetical Monopolist
Paradigm”, US DOJ Antitrust Division Economic Analysis Group, Discussion Paper No. 02-8.
Werden, G.J. (2003). “The 1982 Merger Guidelines and the Ascent of the Hypothetical Monopolist
Paradigm”, 71 Antitrust L.J., 253–276.
1. Introduction
*
Both authors are affiliated with the Bundeskartellamt, Bonn. Arno Rasek is Chief Economist,
Sebastian Wismer is Case Officer within the General Policy Division.
While economists often abstract from market definition within their theoretical models,
practitioners need to get at least some notion about the definition of the relevant market.
Market definition helps to identify customer demand and relevant competitors. 6 Market
definition should inform the competitive assessment and organise it. However, market
definition should not be seen as an end in itself, but a first important step that helps to
assess competitive constraints, market power, and the effects of the behavior at stake. 7
Economists often struggle with the binary nature of market definition and the impact it
can have on the antitrust analysis, in particular as the level of certain market power
indicators depends on market definition. Thus the binary concept has been enriched by
more nuanced concepts such as closeness of competition. In general, the competitive
assessment in a certain case and the definition of the relevant market(s) can be seen as
“communicating vessels”. 8 In principle, a narrow market definition often goes along with
an indication of substantial market power, e.g. a high market share, while a wide market
definition tends to suggest little market power. However, such indications should always
be put into perspective and may in certain cases also be refuted or confirmed by other
circumstances, for instance a detailed analysis of closeness of competition, potential
competition or imperfect (fringe) substitution. 9
As multi-sided markets involve distinct groups of customers which may or may not be
attributed to distinct (but interdependent) markets, these principles on the role of market
definition often become even more important in multi-sided markets. In particular, due to
interdependencies between markets, the (stand-alone) value of market definition may
even be more limited than in one-sided markets.
2. One single market vs. separate markets for distinct market sides
As multi-sided markets involve distinct groups of customers, there are in principle two
alternative approaches to capture their specific structure: defining separate markets for
each customer group or defining a single market encompassing all customer groups.
forces identified within these separate markets, it is easy to identify whether the set of
relevant product substitutes/competitors or the geographic scope differ across markets. In
particular, the analysis may illustrate that a platform operator is dominant, but possibly
not on all market ‘sides’. For example, if one customer group predominantly practices
single-homing while another one practices multi-homing, there might be fierce
competition to attract customers from the single-homing group, but little competition for
customers from the multi-homing group. 10 Overall, with separate markets, it seems
relatively unlikely that the analysis will miss any competition issue that evolves on one of
the ‘sides’ of the market.
However, defining separate markets for each customer group may be inappropriate if the
different groups are inseparably linked by a platform interaction, in particular if a
platform’s service necessarily involves all customer groups. Furthermore, the competitive
analysis may be done repeatedly without gaining additional insights if the set and the
relevance of competitors as well as the geographic scope do not differ across market
‘sides’. Moreover, the risk of missing relevant effects driven by interdependencies
between different customer groups such as indirect network effects seems higher with
separate markets. These aspects militate in favour of defining a single market
encompassing all customer groups. 11
In principle, both approaches seem to be in line with the concept of demand-side
substitutability; in particular, defining one single market does not conflict with this
concept as a platform can be understood as a provider of an intermediation service,
serving linked user groups with essentially the same service. All in all, and given the role
of market definition as a tool that supports competitive analysis, neither of the two
approaches seems right or wrong in absolute terms as long as the analysis appropriately
accounts for interdependencies –such as indirect network effects– and for all competitive
forces on each ‘side’ of the market.
advertisers want to attract the readers’ attention. Consequently, it is not always necessary
for non-transaction platforms to bring both groups of users on board, as some of these
platforms could also exist without one of the two groups. Establishing such non-
transaction platforms can therefore be understood as a strategic business decision of a
firm that would also serve its purpose with only one of the customer groups. 14 All in all,
this suggests defining one single market in the case of a transaction platform while
defining distinct markets in the case of a non-transaction platform. 15
Another similar distinction may be made between “matching platforms” and “audience
providing/advertising platforms”. 16 A matching platform can be described by its objective
to enable the best possible match between different user groups. This objective is shared
by all user groups involved. Although this characterisation partly overlaps with the
definition of a transaction platform, a matching platform may also enable interactions
which do not necessarily imply a subsequent (observable) transaction between user
groups. One example of this type are dating platforms. Although certain matching
platforms also exhibit (negative) direct network effects, 17 they always have positive
bilateral indirect network effects. Hence, transaction platforms can be seen as a sub-
category of matching platforms. In contrast, audience providing platforms or advertising
platforms provide one user group, e.g. advertisers, with the audience or attention of
another user group, e.g. readers. The platform facilitates an interaction between users and
advertisers in the form of a subsequent contact resulting from users reacting to the
advertisement (for instance, by clicking on the ad). Although there might be a certain
matching process involved, the characteristic indirect network effect is unidirectional,
benefiting the advertisers. All in all, this suggests defining one single market in the case
of a matching platform while defining distinct markets in the other cases.
Along with these potential guidelines, it can be useful to investigate the role of the
platform in detail –notably, the extent to which the platform is involved in the interaction
that it enables. On the one hand, this may involve legal questions such as whether the
operator acts as a commission agent or trade representative or bears a substantial part of
specific risks; under certain circumstances these issues are connected with further
questions as to the applicability of specific competition law provisions or, in particular,
the Vertical Block Exemption Regulation. 18 On the other hand, this may lead to
conceptual questions such as whether it is more appropriate to interpret certain market
structures as vertical (upstream and downstream market) rather than two sides of a
platform. 19 However, certain aspects arising in vertical structures, e.g. demand for a wide
range of products within wholesale or retail markets, can have similar implications as
indirect network effects have within multi-sided markets.
Case examples
In Germany, the Bundeskartellamt has identified newspapers as well as magazines as
platforms, i.e. firms that operate in a multi-sided market. However, it has defined two
distinct antitrust markets for readers and advertisers. 20 This seems reasonable since
newspapers and magazines usually do not enable a direct transaction between readers and
advertisers, as they do not necessarily need to get advertisers ‘on board’ to serve readers,
and as the products considered as substitutes usually differ between readers and
advertisers. In contrast, in the case of a merger of two online real estate platforms, the
Bundeskartellamt tended towards defining a single market including both customer
groups, although it ultimately left the market definition open. 21 In a merger decision
concerning online dating platforms, the Bundeskartellamt explicitly defined a common
market including both user groups that are matched by a dating platform. 22 In its decision
Free-of-charge services
In multi-sided markets it can be frequently observed that the platform operator charges
only one customer group while the service is offered for free to another customer group.
There has been some debate as to whether free-of-charge antitrust markets should be
defined. In Germany, the Düsseldorf Higher Regional Court even held that such markets
cannot ‘exist’ in antitrust terms 26 which caused a legislative clarification. 27 It is true that
where there are payments between a supplier and a customer there always exists an
antitrust market. But the inverse conclusion should not be drawn.
Irrespective of whether one single market or separate markets are defined, services
offered free of charge should be considered as (part of) an antitrust market if there exist
indirect network effects between the group that is served without being charged and
another group that is charged. 28 When ignoring one side of a multi-sided market,
important competitive aspects might be missed, as there usually is competition for
customers no matter whether they are paying customers or not. In fact, a customer group
being not charged might be due to intense competition for these customers. However, the
fact that a service is offered free of charge on its own should not justify the definition of a
separate market, in particular as the (zero) pricing decision may reflect both competition
and network effects, and, hence, may be associated with the strategic pricing decision
towards other customer groups. Consequently, when both paid and free-of-charge
services are offered in parallel, it seems reasonable to consider free-of-charge services as
competing services instead of ignoring them.
The approach proposed here also offers a straight-forward answer to the currently
intensely debated question of whether data should be viewed as a ‘currency’ in the
context of internet platforms: 29 for a free-of-charge antitrust market to ‘exist’ it should
not be a requirement that it must essentially be a bundle that comprises a good with a
positive value for the customers (i.e. the platform service) and a good with a negative
value for the customers (i.e. ads, use of their data) which can be viewed akin to a
‘payment’ for the platform service. The reason is that in multi-sided markets, setting a
price of zero for one customer group may make perfect sense for the platform provider
also if the service does not come along with any negative good tied to it. Instead, the
relevant question for the platform provider is to what extent he can monetise the presence
of these customers on other market sides. For the purposes of market definition for
internet platforms, there should thus be no need for the agency to establish that providing
data is of negative value to customers or to even quantify this negative value. As free-of-
charge markets may be defined due to the existence of a different customer group being
charged, there is no need to find a ‘currency’ from the viewpoint of the customers that are
not being charged.
Summarising remarks
Defining one single market seems reasonable for services which mainly aim at enabling a
direct (observable) transaction between different groups, e.g. in the case of a trading
platform that brings together sellers and buyers. In particular, this approach seems
feasible if (i) a firm’s service necessarily involves all groups and (ii) the set of substitutes
and their respective relevance from the perspective of each customer group does not
differ significantly across groups. Otherwise, in particular if the products or services
considered as substitutes (and, hence, competition conditions) differ substantially across
groups, defining a separate market for each distinct customer group seems more
appropriate; in these cases, the resulting markets usually differ in product and/or
geographic scope. These constellations are more likely to exist in cases with non-
transaction or audience providing/advertising platforms. However, market definition and
the choice between the two approaches need to be done on a case-by-case basis.
While the previous section focused on the question of whether separate antitrust markets
should be defined for different sides of a multi-sided market, the following section deals
with the question of whether two platforms belong to the same product market(s) or not.
In principle, the factors relevant for product market definition in single-sided markets
equally apply to multi-sided markets. However, there is a specific phenomenon (more)
frequently found in multi-sided markets that may have significant impact on the antitrust
analysis. In multi-sided markets, pricing and market outcomes depend, among other
things, on whether customers choose a single platform (single-homing) or use more than
one platform simultaneously (multi-homing). In particular, a relatively high degree of
multi-homing within a group of customers may indicate a low level of competition for
these customers, while a relatively high degree of single-homing within a customer group
may indicate intense competition for those customers. 30
In some cases multi-homing can indicate that customers use different platforms in parallel
to cover different needs, even though the platforms’ services may be similar at first view.
For example, in its decision concerning the merger of Microsoft and LinkedIn, the
European Commission distinguished between professional and personal social networks,
in particular because they are used for different purposes and in different ways, although
the technical functionalities of both types of social networks feature several similarities.32
In practice, it is often possible for a competition agency to gain insights on the extent of
multi-homing. However, it might be challenging to interpret this information. Multi-
homing may be a factor mitigating the probability of ‘tipping’ if the two platforms are
substitutes. Multi-homing also tends to reduce the relevance of indirect network effects: if
all customers of one group are present on all platforms, the number of these customers
does not affect the choice between platforms made by members of other groups. 33 Multi-
homing may, however, also indicate that the platforms are not (direct) competitors, while
multi-homing figures alone do not tell us anything about substitutability.
Although the literature on multi-sided markets analyses the impact of multi-homing on
platforms’ decisions and market outcomes in several facets, there seem to be no
contributions that focus on the implications of multi-homing on market definition. Where
one or several customer groups practice multi-homing, agencies should try to investigate
the customers’ multi-homing rationales and consider further splitting of the market, thus
segregating platforms that are used for different purposes and, hence, are not direct
competitors.
Summarising remarks
Customers’ single-homing and multi-homing behaviour can be relevant for market
definition. Much will depend on the underlying rationales. Multi-homing and single-
homing may both justify narrowly defined markets, but the rationale for defining markets
In the following, we will illustrate several challenges as well as peculiarities that arise
when applying traditional methods for market definition in multi-sided markets. The first
part deals with the SSNIP test as a widespread framework which, however, seems
difficult to apply in practice in multi-sided markets. The second part covers some other
quantitative methods, while the third part addresses the role of qualitative evidence.
SSNIP test
One concept that can assist in market definition is the so called SSNIP test. The SSNIP
test was originally developed for one-sided markets. 36 However, due to demanding data
requirements and serious operationalisation issues, the concept should rather be viewed as
an analytical framework as opposed to an easily quantifiable ‘test’.
The original SSNIP test does not account for interdependencies between distinct customer
groups. In a two-sided market, for example, a price increase for one customer group (side
A) leads to changes in demand not only on this side, A, but also on the other side, B.
Ignoring such volume changes that emanate from indirect network effects may distort the
result of the SSNIP test. 37 In case of multilateral positive indirect network effects the
profitability of a price increase would be overestimated, suggesting ‘too narrow’ markets.
Furthermore, even when accounting for volume changes caused by indirect network
effects, the profitability of a (unilateral) price increase also depends on whether prices for
other customer groups can be adjusted. 38
Although approaches to modify the SSNIP test to account for indirect network effects can
be found in the literature, 39 the concept remains difficult to use in multi-sided markets. 40
In practice, the main issues include the lack of proper data on a specific industry (while
data requirements are higher in multi-sided markets), handling of free-of-charge services
as well as the identification and operationalisation of competitive dimensions besides the
price (which might be even more relevant in multi-sided markets). In particular,
modelling and measuring network effects is a non-trivial task, but it is crucial for the
analysis of the SSNIP test as a platform’s pricing leeway may be limited by multilateral
positive network effects or increased by negative network effects. While the sign
(positive or negative) can typically be established, possibly by using qualitative evidence,
the strength as well as the shape of network effects seem difficult to quantify.
Furthermore, multi-sided markets may be especially prone to a “cellophane fallacy” due
to concentration tendencies that multi-sided markets may exhibit. Given these problems,
it is not surprising that so far competition authorities do not seem to have applied a
modified version of the SSNIP test that accounts for multi-sidedness. 41
direct effect of a variation of a strategic variable on the respective firm’s demand is likely
to be overestimated, as part of the demand reaction is driven by a feedback effect.42
However, disentangling these effects in a robust way seems difficult in practice, if proper
data are available at all. Data retrievable for the specific market under review will
typically not contain sufficient (observable) variation with regard to the presence of
indirect network effects that would allow for an econometric quantification of indirect
network effects.
Less complex methods that abstract from modelling demand, such as price correlation
analyses, seem to be more easily applicable. However, multi-sidedness may complicate
the interpretation of calculated substitutability indicators, e.g. correlations, as additional
indirect network effects interfere with substitution as a (direct) reaction on a certain
variation, e.g. a price change. Furthermore, the amount of time until indirect network
effects fully unfold a feedback effect may vary, so the analysis may need to comprise
(different) time lags.
Beyond econometric analyses, it is often useful to apply descriptive quantitative methods.
For example, the matching of customer lists of different platforms can be used to
determine the degree and importance of multi-homing or to identify common customers
and their characteristics. Furthermore, it can be helpful to examine the size of customer
groups and the volume of new subscribers/customers over several periods, in particular if
a party submits that pronounced switching has occurred between certain platforms, as this
may also be reflected in the customer structure or group sizes. In addition, similar as in
one-sided markets, determining catchment areas on the basis of customer locations can be
meaningful when defining the geographic market; however, in multi-sided markets
additional insights can be gained from analysing whether indirect network effects depend
on the location of customers from other groups. If advertisers, for example, are
predominantly interested in targeting customers of a platform who are resident in a
certain region, this may lead to a corresponding segmentation of the market by regions,
even if the advertisers themselves may be based in different regions or countries. Results
of such descriptive methods are often helpful, especially when they complement
qualitative evidence.
Qualitative evidence
Qualitative evidence is (more) frequently used by competition authorities. In particular,
tools such as market studies or an assessment of the consumers’ and other competitors’
points of view can be rather helpful for defining the relevant market(s). 43 Moreover,
surveys and internal documents can often be helpful, e.g. in understanding firms’
rationales behind certain strategic (re)actions or identifying the set of competitors that a
firm perceives and monitors.
Customer surveys in one-sided markets involve well-known problems, e.g. answers to
certain questions from competition authorities might sometimes be biased strategically,
and stated preferences might differ from real reactions.44 In multi-sided markets
additional issues may arise. When investigating stated preferences, in particular, an
implicit or explicit assumption on “other things being equal” might be misleading, as the
choice between alternative offers in presence of network effects also depends on the
choices of other customers. For example, when asking customers about their hypothetical
reaction to a price increase, they may respond to such a question under the (wrong)
implicit assumption that the price increase will not induce any other customer to leave the
platform. Hence, on the one hand it can be useful to assess how important network effects
are for the choices of each customer group, but on the other hand questions concerning
the (hypothetical) substitutability of offers become complicated when both product
characteristics (including price) and network effects drive respondents’ real choices.
Summarising remarks
Competition authorities frequently face the challenge of choosing among investigation
tools which exhibit different strengths and weaknesses and differ in their resource
requirements as well as their reliability. In many cases, authorities refrain from applying
complex econometric methods, in particular due to time constraints, lack of proper data or
methodical complexity which often comes along with limited robustness and difficulties
in interpreting and communicating results.
In multi-sided markets, the analytical complexity is higher if compared to markets
without network effects. Consequently, it seems natural to lean towards simple tools with
a lower degree of complexity. The extent and impact of network effects on both platforms
and their customers should be assessed (at least) qualitatively, in particular to mitigate the
risk of misinterpreting results from established ‘one-sided’ tools.
5. Conclusion
Notes
1 Cf. e.g. Evans and Schmalensee, ‘The Antitrust Analysis of Multi-Sided Platform Businesses’ in
Blair and Sokol (eds), The Oxford Handbook of International Antitrust Economics, vol. 1 (New
York, Oxford University Press 2015), and OECD, ‘Two-Sided Markets’ (2009) Best Practice
Roundtables on Competition Policy.
2 Cf. e.g. Funke/Springer Programmzeitschriften (Case B6-98/13) Bundeskartellamt Decision 25
April 2014, para 138; Zeitungsverlag Schwäbisch Hall (Case B6-150/08) Bundeskartellamt
Decision 21 April 2009, para 33; Visa MIF (Case AT.39398) Commission Decision 26 February
2014 OJ C 168, para 16.
3 Cf. e.g. Wright, ‘One-sided logic in two-sided markets’ (2004) 3:1 Review of Network
Economics; Schiff, ‘The ‘waterbed’ effect and price regulation’ (2008) 7:3 Review of Network
Economics; King, ‘Two-sided markets’ (2013) 46:2 The Australian Economic Review 247-258.
4 Cf. e.g. Hagiu and Wright, ‘Multi-sided platforms’ (2015) 43 International Journal of Industrial
Organization 162-174.
5 Also cf. Rysman, ‘The Economics of Two-Sided Markets’ (2009) 23:3 Journal of Economic
Perspectives 125-143.
6 Also cf. OECD, ‘Market Definition’ (2012) Best Practice Roundtables on Competition Policy.
7 Cf. e.g. ICN Merger Guidelines Workbook, April 2006, p. 15.
8 Cf. e.g. Ewald, ‘Der SIEC-Test im deutschen Recht: Grundansatz, materielle Detailfragen und
praktische Auswirkungen aus ökonomischer Sicht’ (2014) 3 Wirtschaft und Wettbewerb 261-281.
9 Also cf. Bundeskartellamt, ’Guidance on Substantive Merger Control’ (2012) paras 77-78.
10 Cf. e.g. Armstrong, ‘Competition in two-sided markets’ (2006) 37:3 RAND Journal of Economics
680.
11 Also cf. Evans and Noel, ‘The analysis of mergers that involve multisided platform businesses’
(2008) 4:3 Journal of Competition Law and Economics 674, and Monopolies Commission,
‘Competition policy: The challenge of digital markets’ (2015), Special Report No 68, para 58.
12 Filistrucchi, Geradin, van Damme and Affeldt, ‘Market definition in two-sided markets: Theory
and practice’ (2014) 10:2 Journal of Competition Law and Economics 293-339; Luchetta, ‘Is the
Google platform a two-sided market’ (2014) 10:1 Journal of Competition Law and Economics
185-207.
13 Observability (or, more precisely, verifiability) facilitates the platform charging transaction-based
tariffs, extending the space of feasible contracts.
14 Luchetta, ‘Is the Google platform a two-sided market’ (2014) 10:1 Journal of Competition Law
and Economics 192.
15 One may also argue that a distinction based on the platforms’ actual tariff system(s) should be
made; in the case of purely transaction-based fees, defining separate markets might be less
reasonable than defining a single market, cf. Wright, ‘One-sided logic in two-sided markets’
(2004) 3:1 Review of Network Economics 62.
16 Cf. Bundeskartellamt, B6-113/15, Working Paper – The Market Power of Platforms and
Networks, June 2016.
17 Cf. e.g. Goos, van Cayseele and Willekens, ‘Platform pricing in matching markets’ (2013) 12:4
Review of Network Economics 437-457.
18 Cf. e.g. CTS Eventim/FKP Scorpio (Case B6-53/16) Bundeskartellamt Decision 3 January 2017,
paras 101-122.
19 Cf. Hagiu and Wright, ‘Multi-sided platforms’ (2015) International Journal of Industrial
Organization 43 162-174, and Hagiu and Wright, ‘Marketplace or reseller?’ (2015) 61:1
Management Science 184-203.
20 See fn. 2.
21 Immonet/Immowelt (Case B6-39/15) Bundeskartellamt Decision 20 April 2015, case summary
available at www.bundeskartellamt.de.
22 Parship/Elitepartner (Case B6-57/15) Bundeskartellamt Decision 22 October 2015, paras 71-79.
23 CTS Eventim/FKP Scorpio (Case B6-53/16) Bundeskartellamt Decision 3 January 2017, paras
101-122.
24 Google/DoubleClick (Case COMP/M.4731) Commission Decision 11 March 2008 OJ C 184;
Microsoft/Yahoo (Case COMP/M.5727) Commission Decision 18 February 2010 OJ C 020;
Microsoft/Skype (Case COMP/M.6281) Commission Decision 7 October 2011 OJ C 341;
Google/Motorola Mobility (Case COMP/M.6381) Commission Decision 13 February 2012 OJ C
75; Facebook/WhatsApp (Case COMP/M.7217) Commission Decision 3 October 2014 OJ C 417.
25 Travelport/Worldspan (Case COMP/M.4523) Commission Decision 21 August 2007 OJ L 314.
For further case examples, see e.g. Filistrucchi, Geradin, van Damme and Affeldt, ‘Market
definition in two-sided markets: Theory and practice’ (2014) 10:2 Journal of Competition Law
and Economics 293-339.
26 HRS (Case VI Kart 1/14 (V)), Düsseldorf Higher Regional Court 9 January 2015, para 43.
27 In March 2017, the German Parliament passed the Federal Government Bill on the Ninth
Amendment of the German Competition Act; § 18 para. 2a of this Bill explicitly clarifies that
services’ being offered free-of-charge does not conflict with defining an antitrust market.
28 This approach is also in line with the practice of the European Commission that dealt with several
markets including services without charge, cf. fn. 24.
29 Cf. e.g. Schepp and Wambach, ‘On big data and its relevance for market power assessment‘
(2016) Journal of European Competition Law and Practice 7:2 120-124; Körber, ‘Analoges
Kartellrecht für digitale Märkte?’ (2015) 2 Wirtschaft und Wettbewerb 120-133.
30 Cf. e.g. Rochet and Tirole, ‘Two-sided markets: a progress report’ (2006) 37:3 RAND Journal of
Economics 645-667, or Armstrong, ‘Competition in two-sided markets’ (2006) 37:3 RAND
Journal of Economics 668-691.
31 Cf. e.g. Travelport/Worldspan (Case COMP/M.4523) Commission Decision 21 August 2007 OJ L
314 para 13-20.
32 Microsoft/LinkedIn (Case Comp/M.8124) Commission Decision 6 December 2016 para 103-110.
However, it seems that the Commission did not consider multi-homing within the context of
market definition in its decision.
33 Cf. Travelport/Worldspan (Case COMP/M.4523) Commission Decision 21 August 2007 OJ L
314 para 79-80.
34 Also cf. Travelport/Worldspan (Case COMP/M.4523) Commission Decision 21 August 2007 OJ
L 314 para 77-81, and Armstrong, ‘Competition in two-sided markets’ (2006) 37:3 RAND Journal
of Economics 668-691.
35 Cf. e.g. European Commission, ‘Explanatory Note accompanying the Commission
Recommendation on relevant product and service markets within the electronic communications
sector’ SWD(2014) 298, p.28.
36 Starting from a very narrow candidate market, the test asks whether a small but significant and
non-transitory increase in price (“SSNIP”) would be profitable from the perspective of a
hypothetical monopolist in the candidate market. If a SSNIP is not profitable, there probably
exists at least one further relevant substitute product which has not be taken into account. In this
case, it is suggested that the candidate market be expanded until a SSNIP will be profitable from
the perspective of a hypothetical monopolist.
37 Cf. e.g. Evans and Noel, ‘The analysis of mergers that involve multisided platform businesses’
(2008) 4:3 Journal of Competition Law and Economics 663-695, Evans and Noel, ‘Defining
Antitrust Markets When Firms Operate Two-Sided Platforms’ (2005) 3 Columbia Business Law
Review 667-702.
38 Cf. Filistrucchi, Klein & Michielsen, ‘Assessing unilateral merger effects in a two-sided market:
an application to the Dutch daily newspaper market’ (2012) 8:2 Journal of Competition Law and
Economics 322, for an exemplary (numerical) illustration of the application of different modified
versions of the SSNIP test with and without optimal adjustment of the two-sided pricing structure.
39 Cf. Filistrucchi, Geradin, van Damme and Affeldt, ‘Market definition in two-sided markets:
Theory and practice’ (2014) 10:2 Journal of Competition Law and Economics 293-339, and Evans
and Noel, ‘The analysis of mergers that involve multisided platform businesses’ (2008) 4:3
Journal of Competition Law and Economics 663-695.
40 Cf. e.g. Dewenter, Rösch and Terschüren, ‘Abgrenzung zweiseitiger Märkte am Beispiel von
Internetsuchmaschinen‘ (2014) 2 NZKart Neue Zeitschrift für Kartellrecht 387-394, Kehder,
‘Konzepte und Methoden der Marktabgrenzung und ihre Anwendung auf zweiseitige Märkte‘
(Nomos, Baden-Baden 2013), and Haucap and Stühmeier, ‘Competition and antitrust in Internet
markets‘ in Bauer and Latzer (eds), Handbook on the Economics of the Internet (Cheltenham,
Edward Elgar Publishing 2016).
41 Also cf. Filistrucchi, Geradin, van Damme and Affeldt, ‘Market definition in two-sided markets:
Theory and practice’ (2014) 10:2 Journal of Competition Law and Economics 339, and Kehder,
‘Konzepte und Methoden der Marktabgrenzung und ihre Anwendung auf zweiseitige Märkte‘
(Nomos, Baden-Baden 2013) 85-86.
42 Cf. Argentesi and Ivaldi, ‘Market Definition in Printed Media Industry: Theory and Practice’
(2005) CEPR Discussion Paper No. 5096, for an exemplary indication of a bias that may arise
from ignoring two-sidedness when estimating own-price elasticities.
43 Cf. Judgment of the General Court, Topps Europe v Commission, 11 January 2017, T-699/14,
EU:T:2017:2, para 82.
44 Cf. e.g. Davis and Garcés, ‘Quantitative Techniques for Competition and Antitrust Analysis’
(Princeton, Princeton University Press, 2009) 193ff.
45 Also cf. fn. 8.
46 Also cf. Ducci, ‘Out-of-market efficiencies, two-sided platforms, and consumer welfare: a legal
and economic analysis’ (2016) 12:3 Journal of Competition Law and Economics 610.
1. Introduction
This short paper was submitted to the Hearing on "Rethinking the Use of Traditional
Antitrust Enforcement Tools in Multi-Sided Markets", that was held by the OECD
Competition Committee on 22nd June 2017 in Paris. The submission focuses on the topic
of “measuring market power in multi-sided markets”. It is intended to provide practical
and pragmatic suggestions for economists in competition authorities. The paper draws
operational conclusions on how to adapt existing enforcement and merger assessment
tools to address some of the challenges posed by multi-sided markets.
The first section of the paper sets out some important features of multi-sided markets,
including indirect network externalities, single-homing and multi-homing, price structure
and tipping. The second section provides some practical steps in assessing market power
in multi-sided markets and the final section sets out some measures of market power, and
how they may need adaptation in multi-sided markets.
Multi-sided markets are platforms that match two or more groups of customers. Evans
and Schmalensee (2007) define multi-sided platforms as having (a) two or more groups of
customers; (b) who need each other in some way; (c) but who cannot capture the value
from their mutual attraction on their own; and (d) rely on the catalyst of the platform to
facilitate value creating interactions between them.
This section sets out some key features of multi-sided markets that may be important to
an assessment of market power.
1
Kate Collyer is the Deputy Chief Economic Adviser of the Competition and Markets Authority
(“CMA”) in the United Kingdom. Natalie Timan is Director of Economics at the CMA. Hugh
Mullan is Assistant Director of Economics at the CMA. The views expressed are personal to the
authors and all errors, omissions, and opinions are their own
each direction. When there are strong INE in both directions, the interaction between
these INE on both sides can create a feedback loop that may have second and third and
fourth order effects. For instance, the ultimate effect of a price increase to one side of the
market could be much greater if it led to further feedback loops with participants
increasingly leaving both sides of the market as the market becomes less valuable to each
group of customers. The strength of these feedback loops may constrain the platform’s
market power and should be taken into account in any assessment.
Price structure
In a multi-sided market, the price structure reflects the interlinked demands of the two
groups of consumers and the need to get both sides on board. This often results in
complex pricing where the price to each group of consumers does not reflect the marginal
cost of supplying them.
To see the importance of price structure in multi-sided markets, consider the example of a
platform supplying businesses on one side of the market and consumers on the other side.
Assume that in this example consumers are more sensitive to price than businesses. In
order to get consumers on board, the platform allows them to use the service without
charge, but the businesses pay (a fixed fee and/or commission) to be present on the
platform. The platform needs to set a fee to businesses that ensures their participation and
takes account of the feedback loops between both sides of the market. Fewer businesses
will choose to use the services of the platform at higher prices and this will reduce the
attractiveness of the platform to consumers on the other side of the market etc etc.3
As this example shows, the platform must be able to use the price structure to internalise
the externalities arising from the INE. Platforms will always be able to control the price
structure in markets where the two sides do not transact. However, in markets where the
sides do transact, one side of the market can reflect some of the increased costs of doing
business on the platform in the price charged for transactions. Businesses on one side of
the market may pass-through the fees they are charged by the platform to the consumers
on the other side of the market when transacting with those consumers through the
platform. This may undermine the platform’s price structure and limit its ability to
internalise the externalities by facilitating value creating transactions between the two
sides. For example, when a business passes through platform commissions to consumers,
it will not consider how this may reduce consumers’ demand for the platform’s services,
which then affects the demand of all business customers for the platform’s services. It is
only the platform which can take these externalities into account in its pricing to both
sides of the market.
Therefore, in addition to the complex pricing that can be a feature of multi-sided markets,
it will also be important to consider the degree of pass-through when considering the
extent to which multi-sidedness affects the behaviour of the platform.
Tipping
Network externalities can lead to markets tipping to one, or a few, providers. The
feedback loops that can arise when there are strong INE mean that multi-sided markets
tend to be relatively concentrated. A multi-sided market may be less likely to tip the more
differentiated the offering from competing platforms and the more that customers on one
or more sides multi-home. 4 Scale economies and having a critical mass of consumers
may also be important in determining the concentration of a market with platforms
because they influence their financial viability.
Once a market tips, the joint behaviour of consumers and businesses may mean that the
market power of the platform becomes well-established. It may take considerable
co-ordination by both consumers and businesses to switch to another platform to restore
competition. Such co-ordination may be unlikely in the absence of major technological
changes in the sector. For these reasons, establishing whether there is a ‘first-mover-
advantage’ may be important in identifying current market power and the potential
longevity and sustainability of this market power.
When the multi-sided nature of the market is relevant to assessing market power
This discussion suggests that any assessment of market power in multi-sided markets
should take account of these features. The standard results from one-sided markets do not
apply directly to multi-sided markets and any assessment of market power needs to take
this into account explicitly (as we show below). Many of our standard tools for assessing
market power are more complex to apply in multi-sided markets and may need to be
adapted. At a minimum, this may involve simply taking into account the impact multi-
sidedness has on the platforms’ business strategy and decisions. In the next section, we
suggest some practical steps when considering measuring market power in multi-sided
markets.
In this section, we identify some practical approaches which authorities should consider
when measuring market power in multi-sided markets. We discuss these practical
approaches before going on to identify measures of market power.
Understand the nature of competition and identify the market(s) where market
power relevant to the theory of harm is expected to arise
As a first step, an assessment of market power should start from a solid understanding of
the nature of competition in the market under consideration. It should then proceed with
an analytical framework that takes account of any important features arising from the
multi-sidedness of the market.
When thinking about market power and the effect of the conduct, it is important to
identify clearly the nature of competition, including understanding the extent to which
multi-sidedness with multiple consumer groups and interlinked demand affects market
power. This is most likely to be where there are (strong) INEs. For example, market
power on one side of a market may exacerbate market power on the other side, it may
support conduct on another side of the market, or it could be that the market power and
conduct are within the same market, but the conduct also affects another side of the
market. In addition, in multi-sided markets, competitive constraints on market power may
come directly or indirectly from any and all sides of a competing platform. For example,
if a platform tries to engage in exclusion on one side, a rival may be able to respond with
strategies on the other side. This suggests the need to look at all sides of the market when
assessing market power.
The market power we are interested in also depends on the conduct or agreement that we
are interested in. Therefore, measuring market power will be specific to the conduct under
investigation. It is important, at least from an economics perspective, that market power,
is not considered in isolation from the conduct and the theory of harm. 5
In this section, we focus on identifying different measures of market power and explain
how these relate to the conduct considered. These measures of market power are not
exclusive to multi-sided markets. However, we explain how they may need to be adapted
when used in multi-sided markets and we identify some additional challenges that may
arise in this context and where care will need to be taken when interpreting the results of
standard measures. 6
Any assessment of market power should be based on a thorough assessment of the
competitive constraints and in multi-sided markets it will often be necessary to use
multiple sources of evidence and always consider the linked nature of demand.
market shares on one side of the market if the evidence suggests, for example, that that
side is prone to single-homing.
As with all markets, it will be necessary to think through which shares one wishes to
measure. For example, it will not be possible to compute value shares on both sides if one
side does not pay for using the platform. It may then be necessary to measure the number
or value of transactions to calculate market shares. The standard problem of interpretation
also arises with, for example, concerns regarding the relevance of market shares as
measures of market power in markets where services/products are differentiated.
In multi-sided markets, it may be challenging to distinguish between customers and
competitors because customers on one side of the market may also be competitors to the
platform. For example, hotels that list on an online travel agent platform might also
compete directly for bookings. To take another example, third party sellers are customers
on Amazon Marketplace and might also compete with Marketplace to attract direct sales.
Care will be needed to ensure that customers and competitors are correctly identified and
captured in measures of market shares.
Authorities typically aim to identify longer term measures of market power (e.g.
sustained high levels of market share) rather than measures which take a snapshot of a
market in flux or out of equilibrium. However, a multi-sided market with network
externalities may be prone to tipping and authorities may wish to intervene earlier. In that
context, care will be needed to identify whether indications of market power at a
relatively early stage in the development of the market may lead to long term market
power.
The challenges outlined above indicate that care needs to be taken when interpreting what
market shares and, more generally, concentration, indicates about market power in multi-
sided markets.
Single-homing vs multi-homing
The extent to which customers on one side of the market single- or multi-home affects the
single-or multi-homing choice of customers on the other side of the market. Examining
the extent of single or multi homing on each side can provide an indication of likely
market power on each side.
Businesses will benefit from listing on more than one platform if they can play-off the
platforms against each other or if listing on more than one platform expands the number
of consumers in aggregate. For example, a platform may be good at bringing consumers
to the market who would otherwise not participate. If, on the other hand some consumers
single home to platform A and others single home to platform B, then businesses will find
it necessary to use both platforms to reach both sets of consumers). However, single-
homing by different groups of consumers, and multi-homing by none, can lead to market
power for each platform. 7
In markets where INEs are strong it will be important to measure the extent of single or
multi-homing on each side of the market before considering any feedback loops. In
practice, this can be done by gathering information on the following questions:
Conduct
Sometimes the ability to engage in the conduct may be seen as an indicator of market
power, particularly for conduct that would be unachievable or unprofitable in the absence
of market power. 9
Clearly an important factor to consider is how the conduct may lead a market to tip when
a market is already prone to tipping due to the INE.
Switching costs may arise between platforms, or between platforms and direct sales, due
to customer habits and convenience. For example, cookies used by the platform may
mean that it is likely to show a consumer a selection closer to the consumer’s preferences.
The platform may hold the consumer’s payment card details, meaning that these do not
need to be re-entered every time a purchase is made. The platform has the contact details
of the consumer and knows other personal information, so that the platform can contact
the consumer with targeted promotions. Also, the nature of platforms is to reduce search
costs and aid comparability. Therefore, consumers may be expected to prefer this to direct
search across businesses’ own websites.
Technological developments may weaken switching costs as they may lead to periods of
intense innovation and businesses responding to technological changes, which can be
destabilising to established market power. On the other hand, technological developments
may also enhance market power. For example, consumers may be less willing to shop
around through organic browser searches when they have a convenient app on their
phone. Moreover, consumers may not be willing to have numerous apps on their phones
supporting similar services.
5. Assessing the strength and impact of indirect network externalities and feedback
loops
In this final section, we provide practical suggestions for assessing the strength and
impact of indirect network externalities and feedback loops. We have proposed a
sequential approach, looking first at the market power on each side of the market
separately, and second looking at constraints from the other side via the feedback loops.
This second step requires us to assess the strength of feedback loops to examine whether
competition from one side of the market constrains the platform in its price setting to the
other side of the market. This will help establish whether market power on one side of the
market exacerbates market power on another side or whether competition from one side
might constrain the other.
This second step is important because in the presence of strong INE simple one-sided
measures of market power potentially underestimate the market power of the platform.
For example, if the conduct in question undermined the ability of other platforms to
compete effectively, then the presence of strong INE could lead to rapid concentration of
the market and the exclusion of rivals. In this example, if the conduct leads to single-
homing customers on one side of the market switching, the INE may simultaneously act
to strengthen one competitor rapidly and weaken another rapidly. This could be the case
even though static market shares, or other measures, may not indicate a position of
significant market power or dominance.
It is also important to recognise that the potential benefits that a platform may gain from
additional customers on one (or more) side(s) of the market may not always be large. The
incremental value of gaining an additional customer is likely to vary depending on the
number of customers already on the platform. Where a platform already has many
potential members of the market on board, adding one additional business will not
increase the value of the platform to the consumer as much as when the platform had
fewer businesses on board. A platform might therefore put less effort into recruiting
customers once it is more mature. This implies that the pricing structure on the platform
is likely to evolve to reflect the benefit to the platform of additional customers and how
this may change with the total number of customers on the platform. 10
There are two key elements of an assessment of the strength and impact of INE and
feedback loops. The first is the elasticity of demand (on all sides), which provides an
indication of the sensitivity of that group of customers to a change in the relative price.
The stronger the reaction to a change in price, the greater the impact of the feedback loop.
The second element is the responsiveness of demand (on all sides) to participation rates
on the other side(s), which provides an indication of how a response from one side of the
market to a change in price will affect demand on the other side of the market.
In some circumstances, it may be possible to assess the strength of the INE by simply
looking at the rate of growth of the platform and considering how growth in one side of
the market appears to give rise to growth in the other side of the market.
In practice, it may be difficult to measure these elements directly. However, the following
are three potential sources of evidence that may provide information on the strength and
impact of the INE and feedback loops.
• Customer data. If it is possible to collect transaction data for market participants,
it may be possible to use econometric techniques to examine past customer
responses to changes in, for example, platform prices that reveal their preferences.
This data would allow for the direct measurement of both the elasticity of demand
and the responsiveness of demand to participation rates on the other sides. There
are a number challenges with using such evidence, one being that it may be hard
to ascertain the extent to which customers respond by choosing an off-platform
“outside option”.
• Econometric techniques. A combination of evidence on revealed and stated
preference could be used to model choice or estimate demand econometrically. It
may also be possible to measure INE directly using econometric techniques.11 At
present, the theoretical models we are aware of appear to make several
simplifying assumptions and we do not know of any attempts by any competition
authorities to do this. 12
• Survey evidence. Surveys provide a promising source of information on the
strength and impact of feedback loops. Although surveys suffer from the
drawback of using stated preferences, they may have the benefit of not only
providing useful insights into both elasticity of demand and responsiveness of
demand to participation rates, they may also allow for the assessment of
preferences for off-platform options. A survey of businesses, or customers on the
paid side of the market, would allow an authority to gather information on a range
of questions, including: the extent to which the businesses would pass through
increases in the cost of transacting on the platform in the form of higher prices to
consumers on the platform; the value to businesses of consumer participation and
willingness to pay for different rates of participation; the availability of
alternatives and the existence of any switching costs. This could be complemented
with a survey of customers on the other side(s) of the market (i.e. consumers),
which could include questions on how they would react to changes in the relative
price of transactions on the platform, the value to these consumers of business
participation and how different business participation rates would affect their
willingness to use the platform.
These sources of information are unlikely to provide all the evidence required to assess
the strength and impact of INE and feedback loops. The authority will need to make an
assessment in the round and using multiple sources of evidence, including internal
business documents.
6. Conclusion
Where indirect network externalities are strong, the multi-sided nature of the market will
be relevant to the conduct under investigation. The pragmatic approach of assessing
market power in each side of the market and then taking into account feedback loops will
capture the multi-sided nature of the market and its relevance to the conduct under
investigation, provided that it is possible to assess accurately the feedback loops.
We have suggested several practical ways of measuring market power in the different
sides of the market, taking account of the added complexity and potential biases that arise
in using these measures in multi-sided markets. We have also suggested ways of directly
measuring the feedback loops. However, it will not always be possible to measure the
feedback loops directly. Where this is not possible, thinking through how these loops are
likely to work in practice will provide a good qualitative way of capturing the impact
indirect network effects will have on market power.
This annex provides a short summary of some cases which featured multi-sided markets
and were considered by the CMA (or OFT). They illustrate some of the points which
have been made in the main body of the paper and show how the have been applied in
practice.
With commercial radio stations, advertisers pay radio stations for listeners to hear their
commercials and ultimately to increase sales, and listeners purchase radio broadcasting
content by listening to the commercials.
In Global/GMG the merging parties had argued that commercial radio competes with the
BBC for radio audiences and that this has an indirect impact on advertising revenue of
commercial stations given the two-sided nature of the market. 13 This provides an example
of how competition for one side of the market, listeners, may provide a constraint that
protects the other side of the market, even though this competitor does not compete for
the other side of the market. Here, commercial radio stations may be constrained from
increasing the volume of advertising that they allow on their radio stations or degrading
the quality of their programming, because listeners may then switch to the BBC.
Although the OFT considered it credible that there may be some indirect form of
constraint, there was no merger-specific evidence on the extent of this constraint. 14 In
addition, despite recognising the two-sided nature of the market, the OFT chose to focus
primarily on the overlap in radio advertising rather than the overlap between consumers
(listeners) of radio stations, or any adverse effects which may be faced by consumers due
to the merger.
In Global/GCap, the OFT similarly focused its analysis on whether the merger would lead
to advertisers paying more to reach listeners and/or advertisers would receive reduced
value for the money they spend on adverts. 15 Nevertheless, the assessment also
considered how the merger may have negative or positive effects on listeners and how
this may depend on the two-sided nature of the market.
The OFT identified that a loss of competition due to the merger could lead to lower-
quality programming or innovation levels, for example, less investment in paying for top
DJs, presenters, research into play-lists and listeners tastes, and so forth. The OFT noted
that, due to the INE, an adverse effect on listeners, for example due to a reduction in the
quality of programming, would lead to listeners placing a lower value on radio and, as
listener numbers fell, this would have a negative effect on the value which advertisers
place on radio. In this way, the effects are mutually reinforcing, discouraging the merger
parties from deteriorating their programming.
The OFT also considered listeners being “obliged to pay more for the broadcasting
content they seek by being obliged to listen to incrementally more advertising - which can
be considered an adverse effect based on the reasonable assumption that listeners do not
listen to the radio primarily to hear adverts”. The merging parties submitted that they
could broadcast no more than 13 minutes of adverts per hour because this is the tolerance
band of listeners – too many listeners switch off if the proportion of adverts increases
beyond this to make extra advertising profitable. 16
The OFT considered that it may be necessary to balance harm on one side of the market
against benefits on the other side of the market. That is, an increase in prices that harms
the advertiser side of the market may actually benefit the listener side of the market if it
restricts advertising output (total airtime), to the extent that listeners do not listen to the
radio primarily to hear adverts. 17
The assessment in Global GCap also looked at how the merger may lead to efficiencies
and how those efficiencies could be strengthened by to the two-sided nature of the
market. The OFT considered it credible that the merging parties would seek to reposition
their radio stations to make them more differentiated post-merger and this would benefit
listeners and advertisers. 18 The OFT considered that brand repositioning could potentially
improve programming, leading to more listeners tuning-in and as a result advertisers
would be able to reach more listeners, making radio is more valuable to them. 19
Any benefit that listeners gain from re-positioning would also need to be balanced against
any direct price effect to advertisers from the merger. 20 The OFT took some
encouragement from the theory around positive brand repositioning effects in radio
broadcasting having been validated in empirical economic literature.21 Nevertheless, even
in the economics literature the price effects from brand repositioning can be ambiguous.22
In terms of measuring the potential demand-side efficiencies from brand-repositioning,
the OFT considered evidence from the merging parties showing: (i) instances of brand-
positioning which occurred with previous acquisitions, as demonstrated through case
studies; and (ii) the merging parties’ plans to reposition their brands post-merger; and (iii)
evidence on the value that customers place on repositioning. Advertisers were also
supportive in seeing brand repositioning as a favourable development.
Although the discussion above relates to an assessment of efficiencies, it is important to
realise that these arise out of the INE in multi-sided markets and that the same
considerations and measurement techniques may be applicable to measuring market
power. For example, one may use previous instances of entry, expansion or increases in
concentration to test the strength of INE or to assess market power more directly.
Similarly, it is common to look at parties’ internal documents and to understand their
post-merger plans when assessing INE and market power.
The Epyx case provides an example of how a strong preference for single-homing on one
side of the market, as well as the conduct of the firm, has been used in the assessment of
market power.
The CMA’s dominance case related to Epyx’s vehicle service, maintenance and repair
(SMR) platform. This is a commercially available online platform enabling companies
requiring the service, maintenance and repair of corporate vehicle fleets to procure these
services electronically. It is a two-sided service, designed to facilitate the interaction of
one side of the service (buyers, also referred to as demand-side customers) with the other
side (suppliers, also referred to as supply-side customers). The service offers a one-stop
shop for a wide range of functionality covering a wide range of transaction types.
The CMA found that most demand-side customers would prefer to use one SMR platform
only at a given time when processing SMR transactions because multi-homing brings
increased complexity and operational costs of running multiple systems in parallel. 23 The
CMA also found that the SMR processing choices are demand led and that the suppliers
multi-home in response to the single-homing by buyers. Buyers prefer to single-home, so
suppliers provide services on the platforms that buyers use. 24
The CMA also identified how the network effects in this market may lead to barriers to
entry. Demand-side customers do not see much value in joining an alternative platform
unless enough suppliers are subscribed to it, while supply-side customers will only be
inclined to use platforms that have demand-side customers. Therefore, the costs and lead-
times to build a network on both sides of the market were identified as barriers to entry. 25
Challenges in this market were seen to be the need for any new platform to be tested with
customers and the need for the co-operation of Epyx in preparing for and ultimately
affecting a switch during any transitional period. 26
The challenges faced by any potential entrant due to these barriers to entry were made
particularly difficult by the conduct of Epyx, which the CMA considered to be abusive.
This illustrates how the conduct itself may be relevant to the assessment of dominance.
Epyx’s contracts on the demand-side required customers to make all transactions through
Epyx’s platform. They also required customers to pay a minimum annual fee, even if the
volume-related variable fees fell below this fixed fee. Many of the contracts also required
demand-side customers not to ‘develop, use, market or support the sale’ of any alternative
systems. 27 These provisions prevented demand-side customers from developing their own
alternative systems or sponsoring third parties’ alternative systems. 28
Notes
1
For example, the more businesses that join a platform, then the more consumers find that platform
to be attractive; and the more consumers join a platform, then the more businesses find that
platform to be attractive. In addition, the platform may allow advertisers to promote themselves to
consumers (or businesses, or both), which may be a third side of the market.
2
Firms compete aggressively on the side that uses a single network in order to charge monopoly
prices on the other side that is trying to reach them. Armstrong, Mark. 2006. “Competition in
Two-Sided Markets” The RAND Journal of Economics, 37(3): 668-91. As a result, competition
between platforms can have large price effects on the side of the market that uses a single
platform and little or no effect on the side that uses multiple platforms. Rysman, Marc. 2009. “The
Economics of Two-Sided Markets” Journal of Economic Perspectives – Volume 23, Number 3:
125-143.
3
The platform may operate at a loss-making level for some time while it seeks to build up
participation on both sides of the market.
4
However, as already noted, single-homing by customers on one side of the market but across more
than one platform will tend to lead customers on the other side of the market to multi-home. If
customers on one side increasingly single-home on very few platforms, then this would lead to the
market tipping to these platforms despite customers on the other side of the market multi-homing
across these few platforms. Therefore, it will tend to be the increasing extent of single-homing by
the side of the market with most price elastic demand for the platform’s services which will drive
tipping.
5
Some questions that one might ask include: (i) How does any potential market power arise in a
market that has indirect network effects and aspects of multi-sidedness? (ii) How is the behaviour
under investigation related to the market power in the relevant market? (iii) Are the network effects
and multi-sided nature of the market important to the market power? (iv) Are the network effects and
multi-sided nature of the market important to the behaviour being investigated? (v) Is the behaviour
being investigated important for the network effects in the market (e.g. foreclosure which may lead
to the market tipping permanently or preventing some potentially important innovation).
6
As an aside we note that the cellophane fallacy presents a particular challenge when measuring
market power in multi-sided markets, outside of the context of mergers. This standard problem
may arise in any market because, in the presence of market power, prevailing prices would not
equate to competitive prices and the application of the hypothetical monopolist test to prevailing
prices is likely to lead to the relevant market being defined too broadly (i.e. including products
which are not close substitutes at competitive prices). The extent of this problem is likely to
depend on the conduct being considered. In some contexts it may be possible to identify market
power directly without initially defining a market (e.g. by looking at the relationship between
price and concentration in comparable geographical markets). The difficulties arising with the
cellophane fallacy are not particular to multi-sided markets, but may be more challenging because,
as discussed earlier, the nature of these markets means that price will often have little relationship
with measures of cost on either side of the market. Therefore, assessing a competitive price which
is related to a measure of cost is likely to be more challenging. Nonetheless, while it is important
to recognise these difficulties in assessing conduct, the measures of market power identified below
should still be useful.
7
There is an open question as to whether it makes sense to find all platforms as having market
power. Furthermore, do they have market power in the supply of services to businesses (on one
side of that platform) due to the single-homing of the consumers (on the other side of that
particular platform); or do they have market power in the supply of services to the single-homing
consumers? Finally, potential market power due to consumers single-homing on platforms may
not arise if some/many consumers use tools to search across platforms – effectively multi-homing
without necessarily visiting each platform. For example, metasearch sites used in the online travel
industry would appear to support this form of multi-homing (although they appear to account for a
rather small proportion of bookings).
8
We would expect platforms to collect an array of data internally to monitor how it is performing
against internal targets and against rivals. Therefore, internal documents and management
information collected during the normal course of business are likely to provide useful insights.
9
For example, the use of wide MFNs by some platforms might provide some indication of market
power. On the other hand, it may be that the conduct itself impacts upon other measures of market
power. For example, a wide MFN reduces the incentive of businesses to pass-through a
commission increase into their prices on that platform and, to the extent that it is passed though, it
will be matched on other platforms. This means that the initial ‘feedback loop’, which one might
consider in assessing market power, is no longer operational due to the wide MFN.
10
In other words, at the margin, the strength of the INE is unlikely to remain constant.
11
Through simultaneous demand estimation it may be possible to model demand on all sides of the
market and back out the cross elasticities in order to measure the INEs.
12
See, for example, Song, M (2015) “Estimating platform market power in two-sided markets with
an application to magazine advertising.” Working Paper.
13
The BBC is a public service broadcaster which has numerous radio stations but no advertising on
these stations.
14
https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/77575/Global_G
MG_Radio_Repor_PRINT_t.pdf
15
Completed acquisition by Global Radio UK Limited of GCap Media plc ME/3638/08, 27 August
2008. https://assets.publishing.service.gov.uk/media/555de372ed915d7ae5000094
/Global_GCap.pdf
16
In other words, the assessment considered how negative INE, arising due to listeners disliking
advertising, may protect listeners from an increase in the volume of advertising.
17
In contrast to the mutually reinforcing competitive effects described before, the OFT noted that
these competitive effects, which were initiated on the other side (the advertiser’s side) of the
market were inversely related. Para 31
18
A further demand-side merger efficiency in a two-sided market such as radio can occur as a result
of post-merger product or brand repositioning. The basic proposition is that by changing radio
stations format and/or programming post-merger in a way that benefits listeners (that is, by greater
demographic specialisation by individual radio stations), combined radio stations can achieve a
larger and more focussed total audience. The resulting airtime is therefore more valuable to
advertisers seeking to reach a large, focussed demographic.
19
Para 30.
20
The OFT noted the challenges in estimating the different effects: “it is unclear to the OFT how
much—if at all—listeners value each incremental reduction in advertising below the 13 minute
per-hour threshold, nor does the OFT know the curvature of the relationship between price and
total airtime demanded by advertisers for each relevant station affected by the merger” (para 32).
21
See Steven Berry and Joel Waldfogel 'Do Mergers Increase Product Variety? Evidence from
Radio Broadcasting', Quarterly Journal of Economics, August 2001, pages 1009— 1025, who
show that the effect of radio mergers after the US Telecommunications Act of 1996—which
relaxed radio ownership restrictions to differing extents in different-sized markets, effectively
running experiments on consolidation in markets of different sizes—was to increase the amount of
programming variety relative to the number of stations. Other academic work suggests the same
changes also improved radio stations' performance in the market, implying that format changes by
smaller stations may counter the potential exercise of market power by large radio groups that
acquire a substantial share of a particular audience demographic through merger. See Charles
Romeo and Andrew Dick 'The Effect of Format Changes and Ownership Consolidation on Radio
Station Outcomes', Review of Industrial Organisation, December 2005, pages 351—386.
22
See Amit Gandhi, Luke Froeb, Steven Tschantz and Gregory Werden 'Post-Merger Product
Repositioning', Journal of Industrial Economics, March 2008, pages 49—67, who find that the
merged firm moves its product varieties away from each other to reduce cannibalisation and its
competitors move their product varieties between those of the merged firm. Post-merger
repositioning therefore benefits customers by increasing product variety. However, they also find
that repositioning affects post-merger prices in two countervailing ways: there is upward pressure
on all prices as product varieties spread out but the merged firm's incentives to raise price are
reduced as its product varieties move away from each other (as there is less competition between
them to internalise).
23
Para 2.23
24
Para 2.24
25
Para 2.30
26
Para 2.31
27
Paras 3.11-3.12
28
Para 3.14
By Kurt R. Brekke 1
1. Introduction
Multi-sided markets are markets in which a firm serves two or more distinct groups of
consumers. Classical examples include markets for newspapers (serving readers and
advertisers), credit cards (serving shoppers and merchants), and taxis (serving travellers
and drivers). This kind of markets has been around for decades. However, the importance
of multi-sided markets in the economy has increased tremendously, mainly due to
digitalisation and the rapid growth of online markets. 1 While many of these markets are
offering entirely new products to consumers, they also transform traditional one-sided
markets into multi-sided markets due to new business models often based on advertising
as a key source of income.
A key feature of multi-sided markets is the existence of network externalities between the
different sides (consumer groups) in the market, which are by definition not present in
one-sided markets. Network externalities arise when the utility (or profit) obtained by a
consumer (or firm) of one type depends on the number of consumers (or firms) of the
other types in the market and the different consumer groups cannot internalise these
externalities. While the strength of the externality depends on the size of the network, the
sign of the externality can be positive or negative. In the classical newspaper example, it
is quite clear that readers are imposing a positive externality on advertisers, as they are
also potential buyers of the advertised products. This implies that newspapers with large
circulation are likely to attract more advertising revenues. However, the externality on
readers of advertising can be positive, negative or even zero, depending on how
advertising is affecting readers’ utility. 2
The presence of network externalities between the different consumer groups in multi-
sided markets changes the strategic nature of the market game. This has been well-
documented by the large economic literature that has emerged on multi-sided markets. 3 A
main reason is that network externalities affect demand from the different consumer
groups, which in turn influence the firms’ strategic behavior, including pricing decisions.
In the newspaper market, a higher subscription fee will increase the profit margin on
readership but at the same time reduce advertising revenues due to lower circulation.
Thus, the positive network externality from readers to advertisers constrains newspapers
in setting high prices to readers. Indeed, in many online markets, firms are charging zero
user fees to maximise network effects and thus advertising revenues.
1
Chief Economist in the Norwegian Competition Authority (NCA) and Professor at the Norwegian School
of Economics (NHH).
The growing importance of multi-sided markets in the economy poses a key challenge for
competition authorities. A main reason for this is the lack of appropriate tools for
assessing possible anti-competitive effects of firm behavior in such markets. This has
been clearly demonstrated in recent antitrust cases, including the EU cases against
Google, Microsoft and Facebook. 4 While there have been major developments in antitrust
analysis for traditional one-sided markets, such as price pressure tests in merger cases,
these tools cannot directly be applied to multi-sided markets without any adjustments.
Indeed, the nature and strength of the network externalities in multi-sided markets are
likely to determine the anti-competitive effects of firm behavior in such markets.
Applying tools developed for one-sided markets may therefore lead competition
authorities to make wrong decision, such as stopping beneficial mergers (type 1 error) or
clearing harmful mergers (type 2 error).
The purpose of this paper is to explore recent developments in the economic literature on market
power in multi-sided markets, focusing on practical methods and tools that can be applied by
competition agencies, especially in their assessment of horizontal mergers in such markets. The
paper is organised as follows. Section 2 briefly describes the traditional measures of market
power in one-sided markets and the new developments related to price pressure tests. Section 3
reviews the recent developments in the literature on merger assessment tools for multi-sided
markets, whereas Section 4 discusses how these tools can be implemented in practice by
competition authorities. Section 5 concludes the paper with some policy recommendations.
often warrant scrutiny. Mergers resulting in highly concentrated markets that involve an
increase in the HHI of more than 200 points are presumed to be likely to enhance market
power and will usually be investigated by the competition agencies.
However, the use of HHI as a measure of market power has been heavily criticised in
recent years. First, the foundation of HHI in economic theory is based on Cournot
competition with homogeneous products. In such markets firms sell identical products
and compete in quantities, and the price is established by an "auctioneer" that clears
demand and supply. If these are key characteristics of the industry where the merger takes
place, then the HHI is likely to be an appropriate tool for competition authorities.
However, in most markets firms compete in prices and sell differentiated products, which
implies that the HHI can be misleading as an indicator of possible anti-competitive effects
of the merger.
Second, the use of HHI requires a definition of the relevant market, which is usually done
using a so-called "Small but Significant and Non-transitory Increase in Price" (SSNIP) test.
Following this practice is problematic in differentiated product markets, as any HHI-based
analysis neglects information on the substitutability between products, which is decisive for
measuring market power in such markets. While substitutability between products is a
matter of degree, market definition is conceptually different because it involves a zero/one
decision of whether or not to include a given product in the relevant market.
Third, the HHI, as a measure of market power, is difficult to relate to possible efficiency
gains in, say, a merger case. The reason is simply that HHI is a non-monetary measure,
whereas efficiency gains usually are expressed in monetary terms. While it is possible to
translate changes in HHI into price effects, this requires information about price
elasticities, which usually are difficult to obtain for competition agencies. Moreover, even
if it is possible to translate the HHI in monetary terms, the two above-mentioned critiques
still apply, implying that the comparison with efficiency gains is misleading, as the HHI
does not provide a reliable measure of anti-competitive effects, except for markets
characterised by Cournot competition with homogenous products.
As a response, pricing pressure indices have been proposed as alternative measure for
competition authorities when assessing horizontal mergers involving differentiated
products. The framework is based on Bertrand competition with firms selling
differentiated products. The price pressure indices characterise the unilateral price effects
of a horizontal merger by calculating the post-merger effects of marginal price increases
above the pre-merger level. The idea is that, prior to the merger, if one of the merging
firms raises its price by a small amount above the observed equilibrium price, its profits
remain unchanged. Post-merger, if the merged firm increases the price of one of its
products, some of the lost sales will be recaptured by the second product (which used to
be a competing product). Therefore, this price increase is now profitable and thus likely
to occur in the absence of efficiency gains.
The concept of Upward Pricing Pressure (UPP), recently proposed by Farrell and Shapiro
(2010), is based on the idea that a merger changes the firms’ pricing incentives in two ways:
(i) it creates upward pressure on prices due to the loss of competition between the merging
parties’ products and (ii) it leads to downward pressure on prices caused by merger-related
efficiencies (marginal cost decreases). The difference between these two effects is the UPP.
The UPP measure is derived by evaluating the merging firms’ post-merger first-order
conditions at the optimal pre-merger prices, granting the merging firms an efficiency credit.
Considering a merger between firm 1 and 2 selling differentiated products 1 and 2,
respectively, Farrell and Shapiro (2010) define the UPP on product 1 as follows: 7
pressure test can be useful to competition agencies, especially for mergers where linear
demand can be a reasonable assumption. One can also argue that linear demand implies a
conservative measure as the pass-through rate to consumers is 50% of the price change.
In cases where data allow for demand estimation, competition agencies are in a position
to conduct merger simulations that also account for price responses by outsiders. As
prices usually are strategic complements, accounting for such price responses reinforce
any price effect of horizontal mergers. While merger simulations are highly useful in
predicting true price effects of mergers, they are demanding in terms of data and can be
sensitive to methodological assumptions. This often implies that most competition
agencies are not in a position to make use of these tools given the time constraints in
merger cases. In the proceeding we therefore mainly focus on price pressure tests when
considering measures of market power in two-sided markets.
In this section we explore measures of market power in multi-sided markets that can be
employed by competition agencies. A key question is how the measures developed for
one-sided markets can be adjusted to analyse merger effects in multi-sided markets. As
pointed out in the introduction, multi-sided markets differ from traditional one-sided
markets in that (i) firms serve more than one consumer group and (ii) there exists indirect
network effects across the consumer groups. The vast economic literature that has
emerged on multi-sided markets clearly demonstrates that the presence of network effects
changes firms’ strategic behavior and thus the nature of competition.
However, in absence of network externalities across consumer groups, there is really no
difference between one-sided and multi-sided markets. In this case, the competition
authorities can assess the effects of the merger on the different sides of the market
separately, using the standard tools for one-sided markets, as presented above. Indeed,
this is what has been done by competition authorities in many cases until recently. Below
we will show that the standard tools can be misleading in the presence of network effects,
and present new tools for analysing mergers in multi-sided markets. 11
While the literature on multi-sided markets is vast, there are only a few recent studies
developing operational tools for competition authorities’ assessment of mergers in such
markets. An important contribution is the paper by Affeldt et al. (2013) who extend the
UPP measures to two-sided markets. They show that, due to the two-sidedness, the UPP
measures depend on four sets of diversion ratios that can either be estimated using
market-level demand data or elicited in surveys. In an application, they evaluate a
hypothetical merger in the Dutch daily newspaper market. Their results demonstrate that
it is important to take the two-sidedness of the market into account when evaluating UPP.
Let us briefly present the UPP measured developed by Affeldt et al. (2013) for two-sided
markets. In two-sided markets, firms set two prices, one to each consumer group.
Following their example, newspaper 1 set a price 𝑃𝑃1𝐴𝐴 in the advertising market and price
𝑃𝑃1𝑅𝑅 in the readership market, where each of the prices are affecting newspaper 2 in both
markets. A higher 𝑃𝑃1𝑅𝑅 shifts readers from news-paper 1 to newspaper 2. This makes
newspaper 2 more attractive for advertisers, yielding a shift in advertisers to newspaper 2
from newspaper 1. Moreover, a higher 𝑃𝑃1𝐴𝐴 shifts advertisers from newspaper 1 to
newspaper 2. If consumers dislike (like) ads, this shifts readers to (from) newspaper 1
from (to) newspaper 2. Thus, price changes in multi-sided markets involve direct demand
effects, as in one-sided markets, but importantly also feedback effects across sides
(consumer groups) due to network externalities.
Building on Farrell and Shapiro (2010), Affeldt et al. (2013) derive two UPP conditions
for each firm, one for each side of the market. Considering a merger between newspaper
1 and 2, the UPP condition for newspaper 1 in the readership market is given by
𝑈𝑈𝑈𝑈𝑈𝑈1𝑅𝑅 = (𝑃𝑃2𝑅𝑅 − 𝐶𝐶2𝑅𝑅 )𝐷𝐷12
𝑅𝑅𝑅𝑅
− 𝐸𝐸1𝑅𝑅 𝐶𝐶1𝑅𝑅 + �𝑃𝑃2𝐴𝐴 − 𝐶𝐶2𝐴𝐴 �𝐷𝐷12
𝑅𝑅𝑅𝑅
+ 𝐸𝐸1𝐴𝐴 𝐶𝐶1𝐴𝐴 𝐷𝐷11
𝑅𝑅𝑅𝑅
≥0
where the two first terms are the standard UPP measure for one-sided markets, consisting
of the "upward pricing pressure" based on the value of diverted sales from newspaper 1 to
newspaper 2, (𝑃𝑃2𝑅𝑅 − 𝐶𝐶2𝑅𝑅 )𝐷𝐷12𝑅𝑅𝑅𝑅
, net of the "downward pricing pressure" due to merger-
related cost synergies in the production of newspaper 1, − 𝐸𝐸1𝑅𝑅 𝐶𝐶1𝑅𝑅 . However, it is worth
emphasising that firms in multi-sided markets often set user prices below marginal costs,
𝑃𝑃2𝑅𝑅 < 𝐶𝐶2𝑅𝑅 , in order to capitalise on the network effect in the advertising market. In this
case the first term in the UPP measure would be negative, yielding a downward price
pressure, which is opposite of one-sided markets. 12
The two last terms in the UPP condition capture the network effects in two-sided markets.
The first term �𝑃𝑃2𝐴𝐴 − 𝐶𝐶2𝐴𝐴 �𝐷𝐷12𝑅𝑅𝑅𝑅
is the value of diverted sales from newspaper 1 to
newspaper 2 in the advertising market of an increase in the reader price of newspaper 1,
𝑅𝑅𝑅𝑅
where the diversion ratio 𝐷𝐷12 measures the share of advertisers that switch due to fewer
readers of newspaper 1. This is likely to be positive in the case of newspapers, but
𝑅𝑅𝑅𝑅
generally 𝐷𝐷12 can take any sign depending on the nature of the network externality. This
is an additional effect, not present in one-sided markets, which yields an upward
(downward) pricing pressure if the network externality is positive (negative).
The second term 𝐸𝐸1𝐴𝐴 𝐶𝐶1𝐴𝐴 𝐷𝐷11
𝑅𝑅𝑅𝑅
is the synergy effect in advertising costs for newspaper 1, as a
result of the change in the number of advertisers induced by the increase in the reader price.
For the newspaper market, this term is likely to involve a downward pricing pressure on the
reader price. The reason is that synergies in advertising costs imply a higher profit margin
on advertisers, which makes newspaper 1 more reluctant to increase reader prices, as this
𝑅𝑅𝑅𝑅
lowers circulation and thus demand from advertisers. Thus, the "diversion ratio" 𝐷𝐷11 is
likely to be negative in the case of newspapers, but generally the sign depends on the nature
of the network externalities across the different sides of the market.
Affeldt et al. (2013) derive an equivalent condition for the UPP on the advertising side,
which is
𝑈𝑈𝑈𝑈𝑈𝑈1𝐴𝐴 = �𝑃𝑃2𝐴𝐴 − 𝐶𝐶2𝐴𝐴 �𝐷𝐷12
𝐴𝐴𝐴𝐴
− 𝐸𝐸1𝐴𝐴 𝐶𝐶1𝐴𝐴 + (𝑃𝑃2𝑅𝑅 − 𝐶𝐶2𝑅𝑅 )𝐷𝐷12
𝐴𝐴𝐴𝐴
+ 𝐸𝐸1𝑅𝑅 𝐶𝐶1𝑅𝑅 𝐷𝐷11
𝐴𝐴𝐴𝐴
≥0
As for the previous condition, the two first terms are the standard UPP measures for one-
sided markets. The third term is the value of diverted sales from newspaper 1 to
newspaper 1 on the reader side, resulting from an increase in the advertising price 𝑃𝑃1𝐴𝐴 of
𝐴𝐴𝐴𝐴
newspaper 1. The diversion ratio 𝐷𝐷12 measures the share of readers that switch
newspaper as a result of less advertising in newspaper 1, where the sign depends on
whether readers like or dislike advertising. Notice also that the profit margin on the user
side can be, and often is, negative (𝑃𝑃2𝑅𝑅 < 𝐶𝐶2𝑅𝑅 ), which further complicates the
computation of the UPP condition in multi-sided markets. If the profit margin is negative,
then (𝑃𝑃2𝑅𝑅 − 𝐶𝐶2𝑅𝑅 )𝐷𝐷12
𝐴𝐴𝐴𝐴
is positive (negative) if readers dislike (like) ads, and zero if readers
are indifferent.
The last term 𝐸𝐸1𝑅𝑅 𝐶𝐶1𝑅𝑅 𝐷𝐷11
𝐴𝐴𝐴𝐴
captures merger-related synergies in the news production, where
𝐴𝐴𝐴𝐴
𝐷𝐷11 is the change in the number of readers relative to advertisers. A higher advertising
price 𝑃𝑃1𝐴𝐴 implies less advertisers, which may have an impact on the number of readers,
depending on the nature of the network externality, as explained above. Lower costs in
news production yield a higher (or less negative) profit margin on readership. Thus, if
readers like (dislike) ads, this term implies a downward (upward) pricing pressure on the
advertising price of newspaper 1.
Affeldt et al. (2013) derive also GUPPI measures, which ignore efficiency gains, for two-
sided markets:
𝑃𝑃𝑅𝑅 𝑃𝑃𝐴𝐴
𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺1𝑅𝑅 = 𝑚𝑚2𝑅𝑅 𝐷𝐷12
𝑅𝑅𝑅𝑅 2
𝑃𝑃𝑅𝑅
+ 𝑚𝑚2𝐴𝐴 𝐷𝐷12
𝑅𝑅𝑅𝑅 2
𝑃𝑃𝑅𝑅
,
1 1
𝑃𝑃 𝐴𝐴
𝐴𝐴𝐴𝐴 2
𝑃𝑃2𝑅𝑅
𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺1𝐴𝐴 = 𝑚𝑚2𝐴𝐴 𝐷𝐷12 + 𝑚𝑚2𝑅𝑅 𝐷𝐷12
𝐴𝐴𝐴𝐴
𝑃𝑃1𝐴𝐴 𝑃𝑃1𝐴𝐴
where 𝑚𝑚2𝑅𝑅 and 𝑚𝑚2𝐴𝐴 are the profit margins (in percentage) of newspaper 2 in readership and
advertising markets, respectively. The first term in each of the conditions is the standard
GUPPI measure in one-sided markets, whereas the second term captures the network
externalities across the two sides of the market, as explained above.
A recent paper by Cosnita-Langlais et al. (2017) extends (and modifies) the UPP
measures developed by Affeldt et al. (2013). A key point in their paper is that Affeldt et
al. (2013), when deriving the UPP measures, fail to account for within firm feedback
effects in the pricing on the two sides. More precisely, Cosnita-Langlais et al. (2017)
argue that it is unreasonable to assume that the price on one side (say, advertising price
𝑃𝑃1𝐴𝐴 ) is constant when setting the price on the other side (say, reader price 𝑃𝑃1𝑅𝑅 ). 13 Allowing
for within firm feedback effects across the two sides of the market, they derive modified
versions of the GUPPI formula, though under the assumptions of symmetry and linear
demand
𝑅𝑅𝑅𝑅
𝐷𝐷11 𝑅𝑅𝑅𝑅
𝐷𝐷11
𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺1𝑅𝑅 = 𝑚𝑚2𝑅𝑅 �𝐷𝐷12
𝑅𝑅𝑅𝑅
+ 2
𝐴𝐴𝐴𝐴
𝐷𝐷12 �+ 𝑚𝑚2𝐴𝐴 �𝐷𝐷12
𝑅𝑅𝑅𝑅
+ 2
𝐴𝐴𝐴𝐴
𝐷𝐷12 �,
𝐴𝐴𝐴𝐴 𝐴𝐴𝐴𝐴
𝐷𝐷11 𝐷𝐷11
𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺1𝐴𝐴 = 𝑚𝑚2𝐴𝐴 �𝐷𝐷12
𝐴𝐴𝐴𝐴
+ 𝑅𝑅𝑅𝑅 𝑅𝑅 𝐴𝐴𝐴𝐴
𝐷𝐷 � + 𝑚𝑚2 �𝐷𝐷12 + 𝑅𝑅𝑅𝑅
𝐷𝐷12 �
2 12 2
Notice that the first term inside each bracket is the same as in Affeldt et al. (2013). The
additional effect that is pointed out by Cosnita-Langlais et al. (2017) is represented by the
second term in each of the brackets. As they highlight in their paper, these additonal
effects can imply that a merger leading to a price increase on one (say, advertising) side
of the market may lead to a price reduction on the other (say, reader) side, even if there
are no efficiencies and margins are non-negative. This is not case in Affeldt et al. (2013).
Notice, however, that the set of diversion ratios are the same as for the UPP measures by
Affeldt et al. (2013).
In this section we explore how competition authorities can operationalise the market
power tools described above, and obtain reliable estimates of key parameters in multi-
sided markets. An important feature of the pricing pressure indices is that they are based
on parameters that, in principle, are observable to competition authorities, such as
diversion ratios and profit margins in the pre-merger (today) situation. This is not the case
for cost synergies, where the estimates usually are based on plausible "guesses" of future
merger-related cost savings.
The price pressure indices for two-sided markets suggest that competition authorities
need to (i) look at both sides of the market, as an upward pricing pressure on one side can
imply a downward pricing pressure on the other side, and (ii) obtain estimates for
diversion ratios across sides (readers and advertisers) both within and across the merging
firms (newspaper 1 and 2). Following Affeldt et al. (2013), competition authorities, when
assessing mergers in two-sided markets, have to obtain estimates of the following
diversion ratios for the merging parties:
𝑅𝑅𝑅𝑅 𝐴𝐴𝐴𝐴
1. Across products diversion ratios on each of side of the market: 𝐷𝐷12 and 𝐷𝐷12
𝐴𝐴𝐴𝐴 𝑅𝑅𝑅𝑅
2. Across products and sides diversion ratios: 𝐷𝐷12 and 𝐷𝐷12
𝐴𝐴𝐴𝐴 𝑅𝑅𝑅𝑅
3. Within products but across sides diversion ratios: 𝐷𝐷11 and 𝐷𝐷11
Estimates of the six diversion ratios can be obtain by using market or survey data from
the different consumer groups on each side of the market. To illustrate the importance of
accounting for network externalities in two-sided markets, Affeldt et al. (2013) consider a
hypothetical merger in the Dutch daily newspaper. Using estimates for demand
elasticities, prices and marginal costs based on market data, as derived by Filistrucchi et
al. (2012), they compute different UPP measures. Their exercise demonstrates significant
differences between the UPP measures for one-sided and two-sided markets. In particular,
the merger effect in the advertising market is only detected when allowing for network
externalities in the UPP formula.
However, estimates for demand elasticities and marginal costs are usually not available,
and competition authorities need to collect information on diversion ratios using customer
surveys. In a multi-sided market, the survey would need to be more comprehensive, as
one would need to survey consumer groups on all sides of the market. Moreover, one
need to ask the different consumer groups not only how they would react to a price
increase but also how they would react to a change in participation on the other side. 14 A
further complication is that survey results are sensitive to the design of the survey.
Before concluding, let us briefly describe a merger case in the newspaper market in
Norway that was investigated by the Norwegian Competition Authority (NCA). 15 In late
2011 the NCA assessed a proposed merger between the second and the third largest
media houses in Norway. While the parties had several overlapping activities, the concern
for competition was related to local newspapers in overlapping geographical areas. In the
merger assessment, the NCA examined the effects of the merger in both the reader and
advertising markets. The assessment was based on customer surveys of subscribers and
advertisers in six local newspapers. The samples of readers and advertisers were based on
a randomised selection from the actual customer lists of the newspaper, with the final
sample consisting of 200 subscribers and 25 percent of the advertisers for each of the six
newspapers. Information on the consumer groups’ second choice of newspaper was
collected through telephone surveys, asking the question of which newspaper the
subscribers and advertisers would choose if their first choice did not exist. Table 1
summarises the diversion ratios on the two sides of the market.
Subscribers Advertisers
Telemarks avisa → Varden 60% 84%
Telemark county
Varden → Telemarks avisa 51% 49%
Fredrikstad blad → Demokraten 20% 37%
Ostfold county
Demokraten → Fredrikstad blad 20% 58%
To capture the network externality across the two sides of the newspaper market, the
NCA conducted a survey among the subscribers on how they would respond to more
advertisement in the newspaper. The survey showed that consumers were more or less
indifferent towards advertising, suggesting only a one-way network externality from
readers to advertisers. The latter was not measured. The NCA proceed by considering the
two sides of the market independently, but with a discussion of the network externality
from readers to advertisers. The merger was eventually approved in June 2012, with the
remedy that the parties divested two newspapers, one in each of the local markets.
While this case is an early attempt to account for network externalities of mergers in two-
sided markets, the analysis by the NCA has, in light of the UPP measures described
above, analysis several shortcomings. First, the NCA did not estimate the profit margins,
which is important in two-sided markets. As shown above, if the newspaper profit margin
on the reader side is negative, the network externality effect is likely to impose a
downward pricing pressure on the reader price, whereas the opposite is true if this profit
margin is positive. Second, the NCA did not estimate diversion ratio related to the
network externality from readers to advertisers, which would be a necessary input in the
computation of the UPP measures accounting for the two-sidedness, as shown by Affeldt
et al. (2013).
5. Concluding remarks
In this paper we have reviewed the recent literature on market power measures in multi-
sided market, and based on this described operational tools that can be employed by
competition agencies, especially in the assessment of mergers in such markets. The paper
has focused mainly on the recent developments of pricing pressure indices, which is
probably the most likely tools to be used by most com-petition authorities, as full merger
simulations are quite demanding due to tight time constraints in merger cases. The key
lessons from this review can be summarised as follows:
1. Upward pricing pressure on one side of the market may result in downward
price pressure on the other sides due to network externalities;
2. Upward pricing pressure can be reinforced or weakened depending on the
nature of the network externality, i.e. whether the externality is positive or
negative.
3. In case of one-way network externalities (say, only from readers to
advertisers), then standard UPP measures can be employed on the side that
benefits from network externality (advertising side) but not on the other sides
causing the network externality (reader side).
Thus, using standard market power measures developed for one-sided markets yield
misleading estimates of anti-competitive effects of firm behavior. In the case of mergers,
we have shown that competition agencies cannot assess each side of the market
separately, but need to adopt tools that account for possible network externalities across
the different sides of the market. In particular, we have described recent developments in
the economic literature that suggest modified versions of UPP and GUPPI that can be
adopted by competition agencies.
By way of conclusion, we should stress some limitations with the UPP measures. First,
the general critique that applies to using pricing pressure indices in one-sided markets
remains valid also for multi-sided markets. In particular, the fact that no assumption on
demand systems are needed (which determines pass-through) is because both UPP and
GUPPI only calculate the incentive to increase prices unilaterally post-merger but not the
actual price increase. However, what one is ultimately interested in is the change in total
welfare and consumer surplus due to the merger, which is determined by the merger-
induced price change. 16
Second, the UPP measures ignore responses by competitors. If the merging parties
increase their prices post-merger, competitors have an incentive to also increase their
prices in response. This is turn gives the merging parties the incentive to raise prices
further. Hence, UPP and GUPPI tend to underestimate the incentive to increase prices
post-merger in a one-sided market. In a two-sided market, depending on the sign and size
of the indirect network effects, prices on one side might be strategic complements (as in
one side markets) and strategic substitutes on the other side. Therefore, UPP and GUPPI
may either underestimate or overestimate the incentives to increase prices.
Notes
1 See, for instance, Evans and Schmalensee (2016) who clearly demonstrates the importance new
markets related to multi-sided platforms (matchmakers).
2 See, Kaiser and Wright (2006), Kaiser and Song (2009), and Wilbur (2008), for empirical
evidence on this relationship.
3 See, for instance, Anderson and Jullien (2015) or Evans and Schmalensee (2016).
4 Google/DoubleClick (Case COMP/M.4731) Commission Decision 11 March 2008 OJ C 184;
Microsoft/Yahoo (Case COMP/M.5727) Commission Decision 18 February 2010 OJ C 020;
Microsoft/Skype (Case COMP/M.6281) Commission Decision 7 October 2011 OJ C 341;
Facebook/WhatsApp (Case COMP/M.7217) Commission Decision 3 October 2014 OJ C 417.
5 This is obviously a simplification, as it is well known from the economic literature that both
merging and non-merging firms are likely to change their behaviour as a consequence of the
merger.
6 See U.S. Department of Justice & FTC, Horizontal Merger Guidelines § 5.2 (2010).
7 There is, of course, an equivalent UPP condition for product 2.
8 The diversion ratio is formally defined as follows
𝜕𝜕𝑄𝑄2 /𝜕𝜕𝑃𝑃1
𝐷𝐷12 : = ,
−𝜕𝜕𝑄𝑄1 /𝜕𝜕𝑃𝑃1
where Q1 and Q2 are the demands for product 1 and 2. Thus, the diversion ratio measures the
share of consumers of product 1 that switch to product 2 due to a price increase of product 1.
9 Formally, the merger-related efficiency gain of product 1 is defined as follows:
𝐸𝐸1 : = (𝐶𝐶1 − 𝐶𝐶1𝑁𝑁 ) / 𝐶𝐶1 ,
Where 𝐶𝐶1𝑁𝑁 is the post-merger marginal cost of product 1. It is assumed that 𝐶𝐶1𝑁𝑁 ≤ 𝐶𝐶1 such that
𝐸𝐸1 ∈ [0, 1].
10 Schmalensee (2014) provides an alternative version of the UPP by allowing for also efficiency
gains in the production of both products, yielding the following condition
𝑈𝑈𝑈𝑈𝑈𝑈1 = [𝑃𝑃2 − (1 − 𝐸𝐸2 ) 𝐶𝐶2 ] 𝐷𝐷12 − 𝐸𝐸1 𝐶𝐶1 ≥ 0
𝐸𝐸2 is the merger-related efficiency gain in production of product 2, which evidently increases the
upward pricing pressure by increasing the value of diverted sales.
11 See, for instance, Rochet and Tirole (2006) who derive a modified version of the Lerner index for
two-sided markets.
12 Note, however, that if 𝑃𝑃2𝑅𝑅 < 𝐶𝐶2𝑅𝑅 , this must imply that 𝑃𝑃2𝐴𝐴 > 𝐶𝐶2𝐴𝐴 , otherwise the firm is running
deficits.
13 When deriving the UPP formula, Affeldt et al. (2013) assume that all other prices are constant.
14 This has already been done by competition agencies in some merger cases, there are no example,
to our knowledge, of these being used to compute UPP measures accounting for the network
externalities in multi-sided markets.
15 This case (Case 2011/0925: A-pressen AS – Edda Media AS) is described in OECD report (2016)
on the Roundtable on Market Definition in Two-Sided Markets.
16 See, for instance, Fan (2013) for a full merger simulation in the US newspaper market.
References
Affeldt, P., Filistrucchi, L., Klein, T.H., 2013. Upward pricing pressure in a two-sided market. Economic
Journal, 123, 505-523.
Anderson, S.P., Jullien, B., 2015. The advertising-financed business model in two-sided media markets.
Chapter 2 in Handbook of Media Economics (Eds. Anderson, S., Waldfogel, J., Strömberg, D.),
Elsevier, 41-90.
Argentesti, E., Filistrucchi, L., 2007. Estimating market power in a two-sided market: the case of
newspapers. Journal of Applied Econometrics, 22, 1247-1266.
Cosnita-Langlais, A., Johansen, B.O., Sørgard L., 2016. Upward price pressure in two-sided markets:
incorporating feedback effects. Memo.
Fan, Y., 2013. Ownership consolidation and product characteristics: a study of the U.S. daily newspaper
market. American Economic Review, 103, 1598-1628.
Farrell, J., Shapiro, C., 1990. Horizontal mergers: an equilibrium analysis. American Economic Review,
80, 107-126.
Farrell, J., Shapiro, C., 2010. Antitrust evaluation of horizontal mergers: an economic alternative to
market definition. B.E. Journal of Theoretical Economics, 10, article 9.
Filistrucchi, L., Klein, T.J., Michielsen, T., 2012. Assessing unilateral merger effects in a two-sided
market: an application to the Dutch daily newspaper market. Journal of Competition Law and
Economics, 8, 1–33.
Hausmann, J., Moresi, S., Rainey, M., 2011. Unilateral effects of mergers with general linear demand.
Economics Letters, 111, 119-121.
Kaiser, U., Wright, J., 2006. Price Structure in Two-Sided Markets: Evidence from the Magazine
Industry. International Journal of Industrial Organization, 24, 1-28.
Kaiser, U., Song, M., 2009. Do media consumers really dislike advertising? An empirical assessment of
the role of advertising in print media markets. International Journal of Industrial Organization 27,
292-301.
Moresi, S., 2010. The Use of Upward Pricing Pressures Indices in Merger Analysis. Antitrust Source,
February 2010.
OECD, 2012. Roundtable on market definition: Note by the delegation of Norway.
DAF/COMP/WD(2012)21.
Rochet, J.-C., Tirole, J., 2006. Two-sided markets: a progress report. RAND Journal of Economics, 37,
645-667.
Salop, S.C., Moresi, S., 2009. Updating the merger guidelines: comments. Mimeo, Charles River
Associates, London.
Wilbur, K.C., 2008. A two-sided, empirical model of television advertising and viewing markets.
Marketing Science, 27, 356-378.
By Michael L. Katz 1
1. Introduction
The topic of this paper lies at the intersection of two concepts: multi-sided markets and
exclusionary behaviour. This is a challenging topic for at least two reasons. First, there is
a lack of consensus as to what constitutes a multi-sided market. Second, there is
considerable disagreement about what constitutes exclusionary behaviour —whether or
not one is examining a multi-sided market.
The lack of a consensus definition of multi-sided markets is somewhat easier to address
(or, at least, to hold to one side). Suppliers in multi-sided markets are often referred to as
“platforms” because they serve as bases on which users from different sides of the
markets can interact with one another. For antitrust purposes, a useful definition of a
multi-sided market is that there are cross-platform network effects (i.e. the presence of
members of group A as users on one side of the platform makes the platform more
attractive to members of group B on the other side) in at least one direction for a platform
that facilitates interactions between two or more groups of users, can set distinct prices to
different user groups, and has market power with respect to those groups. 1 This definition
captures the sorts of situations that are commonly labelled as platforms or multi-sided
markets in recent antitrust litigation.
The lack of agreement regarding what constitutes exclusionary behaviour is more
problematical. There is a broad consensus that conduct is exclusionary when it harms the
competitive process by weakening the ability of rival firms to compete and the conduct
does not constitute competing on the merits. However, there is considerable disagreement
regarding what it means to “harm competition” or to fail to “compete on the merits”.
Consequently, the discussion below begins, in Section 2, with an examination of broad
conceptions of exclusion, without focusing specifically on multi-sided markets.
The paper then turns to the question: How should antitrust enforcers and the courts
identify whether the conduct of a firm operating in a multi-sided market is exclusionary? 2
As Rochet and Tirole (2006, p. 646) have observed, multi-sided markets combine
elements of multi-product pricing and network effects. As a result, the issues are not
entirely new or unique, but they are challenging nonetheless. Specifically, multi-product
pricing and network effects raise several issues for competition policy’s treatment of
exclusionary behaviour:
1
Sarin Professor Emeritus in Strategy and Leadership, Haas School of Business, and Professor
Emeritus of Economics, University of California Berkeley
Equally-efficient-rival test
A second test asks whether an equally efficient rival could compete successfully in the
presence of the challenged conduct. If the answer is yes, then by this test the conduct is
not exclusionary. 8 This test builds on an intuitive notion of harming competition under
which, if a firm is competing on the merits, then an equally matched rival should find
itself capable of competing successfully as well. Unfortunately, this approach suffers
from both practical and conceptual shortcomings.
A severe practical shortcoming is that, in actual markets, it can be very difficult to
determine what it means to be an equally efficient rival. When each supplier offers a
single product that is undifferentiated from those of its rivals, the determination is
straightforward: a rival offering the same product to consumers is equally efficient if it
has costs lower or equal to those of the firm in question. However, when products are
differentiated, it is necessary to account for the differences. It can be extremely difficult
to determine whether a competitor is equally efficient when product characteristics and
business strategies are multidimensional and vary across firms. For example, given the
many differences in their business models, it might be very difficult to assess whether
American Express and MasterCard are equally efficient credit and charge card platforms.
In markets with network effects, additional issues arise. Should the size of a rival’s
installed base be taken into account in defining what it means to be equally efficient? If it
is, then there may be a risk that this test will become extremely weak because it would
find any conduct that leveraged a dominant firm’s installed base advantage to be non-
exclusionary regardless of how it affected competition and consumer welfare. However,
not taking installed bases into account might have the effect of forcing a firm with a large
installed base to refrain from competing vigorously with a smaller rival.
In summary, the equally-efficient-rival test can be very hard for the courts to apply, and it
can thus create uncertainty for potential defendants and lead to some of the problems
associated with application of a social-welfare standard as discussed above.
An even deeper shortcoming of the equally-efficient rival test is that its focus on an as-
efficient competitor lacks a sound grounding in economics. Specifically, there is not a
tight linkage between: (a) the consumer and total-welfare effects of competition between
two firms, and (b) whether the two firms are equally efficient suppliers. For example, in
the presence of production economies of scale, the entry of an equally efficient rival can
lead to higher industry costs to produce a given amount of output, and -from the
perspective of total surplus- these higher costs may dominate any benefits of the
additional competition due to entry. A similar problem can arise with network effects,
which give rise to demand-side economies of scale. In the other direction, consumer
surplus will often rise following entry even if the entrant is less efficient than the
incumbent. Indeed, given the effects on prices and consumption, entry by an inefficient
entrant can raise total surplus in some instances.
The equally-efficient-rival test broadly underlies the European Commission’s assessment
of price-based exclusionary behaviour and whether it might give rise to consumer harm. 9
However, the Commission recognises that excluding a less efficient competitor can harm
competition in some circumstances. 10 The Commission also recognises that, in the
presence of network effects, a rival’s efficiency can be affected by exclusionary
conduct. 11
Next consider the definition of exclusion for the specific practice of predatory pricing.
Following the U.S. Supreme Court in Brooke Group, U.S. courts apply a two-part test for
predation. “First, a plaintiff seeking to establish competitive injury resulting from a rival's
low prices must prove that the prices complained of are below an appropriate measure of
its rival's costs.” 15 “The second prerequisite to holding a competitor liable under the
antitrust laws for charging low prices is a demonstration that the competitor had a
reasonable prospect, or ... dangerous probability, of recouping its investment in below
cost prices.” 16 The European Union standard has a multi-band price-cost prong: (a) if
price is below average variable costs, then there is a presumption of predatory pricing that
the defendant can then attempt to rebut, and (b) if price is above average variable cost but
below average total cost, then the plaintiff must establish that the pricing is intended to
eliminate competitors. 17 The European Union standard does not have a required
below-cost pricing can be profitable even for a monopolist facing no threat of entry,
which demonstrates that such pricing can be motivated by considerations other than
exclusion. In addition to benefiting the supplier, this type of pricing can benefit
consumers by internalising what would otherwise be externalities across user cohorts (i.e.
early users do not take into account the benefits of a larger network size that their
purchases confer on later user cohorts). However, as discussed above, a supplier can also
be motivated by an exclusionary desire to deny its rivals the benefits of increasing their
own installed bases. Indeed, both types of incentives can be present simultaneously. 25
The fact that above-cost prices are predatory in some circumstances, and below-cost
prices constitute competition on the merits in others, strongly suggest that there is no
good price-cost test in the presence of network effects. Using a formal model of same-
side network effects with two user cohorts, Farrell and Katz (2005) have shown that price
floors that fully promote total surplus would have to depend on user expectation and
co-ordination processes that are unlikely to be observable in practice. In many respects,
the two user cohorts in a two-period model of same-side network effects play the same
role as the two user groups on opposite side of a platform. 26 Hence, these results strongly
suggest that price-cost test is problematical when applied to a multi-sided platform.
Suppose that, despite the issues inherent in the use of marginal cost as a bright line for
identifying predatory pricing, one attempts to extend the Areeda-Turner price-cost test to
multi-sided markets. Consider a platform that facilitates exchanges between members of
user group A and user group B. A naive application of the Areeda-Turner test might focus
on the pricing to users on one side of the platform, say side A, in isolation. That is, the
price-cost prong would examine whether 𝑝𝑝𝐴𝐴 is less than 𝑐𝑐𝐴𝐴 , where 𝑝𝑝𝐴𝐴 is the price charged
to members of user group A, and 𝑐𝑐𝐴𝐴 is the marginal cost of providing a unit of platform
services to a member of user group A.
As has long been emphasised by contributors to the academic literature on multi-sided
platforms, this naive approach can be highly misleading. 27 To see why, consider a
platform that: (a) facilitates one-to-one transactions; (b) charges fees to users solely on a
per-transaction basis (i.e. it does not charge subscription fees); and (c) incurs only fixed
costs or per-transaction costs (i.e. there are no marginal costs associated with changes in
the number of platform subscribers if ones holds the total number of transactions fixed).
Let 𝑥𝑥𝐽𝐽 denote quantity of platform services consumed by users on side J. For such
platforms, 𝑥𝑥𝐴𝐴 ≡ 𝑥𝑥𝐵𝐵 and there may be no sound basis for assigning costs to one side or
other. Let 𝑐𝑐𝑇𝑇 denote the total marginal costs associated with a transaction. Because costs
are associated with transactions -not one side of the market or the other- and because
transactions only occur if both sides participate, it also makes sense to think of revenues
at the transaction level. That is, the firm earns 𝑝𝑝𝐴𝐴 + 𝑝𝑝𝐵𝐵 per transaction. Applied at the
transaction level, the two-sided market version of the Areeda-Turner test compares
𝑝𝑝𝐴𝐴 + 𝑝𝑝𝐵𝐵 with 𝑐𝑐𝑇𝑇 .
This comparison highlights the fact that a simple, one-sided price can be misleading.
Under the naive approach, policy enforcers would have to assign some share of the total
transactions costs to one side of the market. Let 𝜆𝜆 denote the percentage of the cost of a
transaction allocated by the competition authority to side A. It could well be the case that
the naive, one-sided version of the test indicates below-cost pricing (i.e. 𝑝𝑝𝐴𝐴 − 𝜆𝜆𝑐𝑐𝑇𝑇 < 0)
while the two-sided version does not (i.e. 𝑝𝑝𝐴𝐴 + 𝑝𝑝𝐵𝐵 − 𝑐𝑐𝑇𝑇 > 0). Because the one-sided
version would rely on arbitrary allocations of costs and revenues, it is difficult to see why
it would be preferred to the two-sided version, which examines costs and revenues at the
transaction level.
Another way to see the dangers of focusing solely on one side of a multi-sided market is
to recognise that there is an important sense in which a multi-sided market is no different
than any other -in each case, it is necessary to compare prices and costs. For some
purposes, it is not too much of stretch to consider any firm as a platform that facilitates
transactions between input suppliers and output buyers, where the input suppliers pay
negative prices to participate on the platform. From this perspective, looking at the price
paid by buyers minus the price paid to input owners amounts to taking both sides of the
market into account at once. Moreover, in the presence of network effects, users on one
side of platform can be viewed as inputs to the supply of services to users on the other
side, and the cost of that input has to be taken into account.
Behringer and Filistrucchi (2015) derive the two-sided analog of the Areeda-Turner test
for platforms that are not pure transaction facilitators. One example of this type of
platform is a media company that sells subscriptions to households and advertising to
firms seeking to reach households. A critical point of distinction from the pure-
transactions situation discussed above is that the platform’s unit sales to the two sides of
the market need not be equal to one another (i.e. it may be the case that 𝑥𝑥𝐴𝐴 ≠ 𝑥𝑥𝐵𝐵 .
Although they need not be equal, the unit sales on the two sides of the market will affect
one another when there are cross-platform network effects. It is thus necessary to account
for the fact that an increase in sales on one side of the platform generates costs and
benefits on the other side of the platform.
Behringer and Filistrucchi (2015) consider a monopolist facing demand 𝑥𝑥𝐵𝐵 =
𝑥𝑥𝐵𝐵 (𝑥𝑥𝐴𝐴 , 𝑝𝑝𝐵𝐵 ). In the presence of positive cross-network effects, an increase in 𝑥𝑥𝐴𝐴 leads to
increased demand by side B, holding the price charged to side B constant. Behringer and
Filistrucchi propose a two-sided test under which a necessary but not sufficient condition
for finding predatory pricing is that at least one of the following amounts is negative:
𝜕𝜕𝑥𝑥𝐵𝐵
(𝑝𝑝𝐴𝐴 − 𝑐𝑐𝐴𝐴 ) + (𝑝𝑝𝐵𝐵 − 𝑐𝑐𝐵𝐵 )
𝜕𝜕𝑥𝑥𝐴𝐴
and
𝜕𝜕𝑥𝑥𝐴𝐴
(𝑝𝑝𝐴𝐴 − 𝑐𝑐𝐴𝐴 ) + (𝑝𝑝𝐵𝐵 − 𝑐𝑐𝐵𝐵 ).
𝜕𝜕𝑥𝑥𝐵𝐵
There are several points worth noting about this test. First, as in traditional markets, the
Areeda-Turner rule lacks a tight linkage to welfare. Even using Behringer and
Filistrucchi’s formulas to determine whether prices are above or below costs, there can be
above-cost pricing that lowers welfare by weakening rivals and below-cost pricing that
raises welfare.
Second, these formulas can be interpreted in ways that implement the no-economic sense
test of predation. However, one must be careful about the calculation of the margin and
demand terms in these formulas in order to ensure that one does not count as benefits any
gains that the platform might obtain by reducing the number of users on the other
platform or by inducing that platform to raise its prices.
In order to understand the need for caution with respect to the demand terms, 𝜕𝜕𝑥𝑥𝐵𝐵 ⁄𝜕𝜕𝑥𝑥𝐴𝐴
and 𝜕𝜕𝑥𝑥𝐴𝐴 ⁄𝜕𝜕𝑥𝑥𝐵𝐵 above, it is helpful to expand the notation slightly. Label the platform under
scrutiny by i and a rival platform by –i. Using notation that accounts for the presence of
the competing platform, the demand faced by platform i can be expressed as 𝑥𝑥𝐵𝐵𝑖𝑖 =
𝑥𝑥𝐵𝐵𝑖𝑖 (𝑥𝑥𝐴𝐴𝑖𝑖 , 𝑥𝑥𝐴𝐴−𝑖𝑖 , 𝑝𝑝𝐵𝐵𝑖𝑖 , 𝑝𝑝𝐵𝐵−𝑖𝑖 ). One would expect group-B users’ demand for platform i to fall as
either the rival’s price falls or its group-A user base rises. The demand of users on the
other side of the platform can be defined similarly. The no-economic-sense logic implies
that the appropriate value of 𝜕𝜕𝑥𝑥𝐵𝐵 ⁄𝜕𝜕𝑥𝑥𝐴𝐴 to use for platform i in the pricing formula above
is 𝜕𝜕𝑥𝑥𝐵𝐵𝑖𝑖 ⁄𝜕𝜕𝑥𝑥𝐴𝐴𝑖𝑖 because this term does not represent any weakening of the rival.
It is important to recognise that one cannot estimate 𝜕𝜕𝑥𝑥𝐵𝐵𝑖𝑖 ⁄𝜕𝜕𝑥𝑥𝐴𝐴𝑖𝑖 simply by looking at how
sales to group-B users rise when the platform lowers its price to group-A users and the
number of group-A users rises in response. The reason is that the price change will also
affect the number of group-A users on platform –i. Specifically, by making platform i
more attractive to side-A users, lowering 𝑝𝑝𝐴𝐴𝑖𝑖 will raise 𝑥𝑥𝐴𝐴𝑖𝑖 and lower 𝑥𝑥𝐴𝐴−𝑖𝑖 . Both of these
changes in the numbers of users will raise 𝑥𝑥𝐵𝐵𝑖𝑖 , but only the first effect should be counted
under a no-economic-sense standard; the latter constitutes a weakening of the rival. 28
Another way to see this point is to consider a situation in which there are multiple cohorts
of users over time. As discussed above, a network might charge lower prices to early
cohorts in order to: (a) build up its own installed base to offer greater network benefits to
later cohorts of users, and/or (b) prevent rivals from becoming stronger future competitors
by building up their own installed bases. Adopting a multi-sided perspective, one might
be tempted to take both types of benefits into account because the core of the approach is
to account for the platform’s gains and losses associated with all users (here, different
cohorts), rather than focusing on one group in isolation. But notice that, the more
successful the firm is in weakening rivals (and, thus, generating future sales), the more
this form of the test indicates that the firm is not engaged in predation. Intuitively, this
form of the price-cost test mistakenly treats recoupment as covering costs.
In addition to the demand terms, the price-cost margins must also be interpreted with
care. In some circumstances, charging lower prices to the A side of a market may weaken
competition on the B side and, thus, allow the platform to charge higher prices to B-side
users. Critically, in these circumstances, the higher prices are due to the loss of
competition rather than an increase in cross-platform network effects. A naive test would
count the elevated prices as offsets to the predatory prices rather than recognising them as
a form of recoupment occurring at the same time as the predatory pricing.
A recent case brought by the United Kingdom’s Director General of Fair Trading
illustrates this issue. 29 Napp Pharmaceutical Holdings Limited and subsidiaries sold oral
sustained-release morphine to two market segments: hospital (i.e. patients in hospital) and
community (i.e. patients under the care of a general practitioner). The Director found that,
due to switching costs and reputational effects, purchase decisions of the community
segment were strongly influenced by purchase decisions of the hospital segment. This
influence gave rise to form of cross-platform network effect: all else equal, greater
hospital sales could be expected to lead to greater community sales. Moreover, a supplier
lacking substantial hospital sales would have difficulty effectively competing in the
community segment.
The Director found, in part, that Napp charged predatory, below-cost prices to the
hospital segment in order to prevent entry and weaken competition in the community
segment. In its defense, Napp argued that its prices to the hospital segment were not
predatory because they generated profitable sales in the community segment. Letting A
denote the hospital segment and B the community segment, Napp’s argument can be
stated in terms of the formulas above. Napp’s position was that, even if (𝑝𝑝 𝐴𝐴 − 𝑐𝑐𝐴𝐴 ) < 0,
the prices were justified because
𝜕𝜕𝑥𝑥𝐵𝐵
(𝑝𝑝𝐴𝐴 − 𝑐𝑐𝐴𝐴 ) + (𝑝𝑝𝐵𝐵 − 𝑐𝑐𝐵𝐵 ) > 0.
𝜕𝜕𝑥𝑥𝐴𝐴
The Director argued -and the Competition Appeal Tribunal agreed- that Napp earned
“high compensating margins in the community segment... precisely because its discount
policy in the hospital segment has hindered competition in the community segment.” 30
The Tribunal explained that: 31
the fact that Napp’s below-cost pricing in the hospital sector enables it to make
money from ‘follow-on’ sales in the community sector merely signifies that the
particular form of ‘recoupment’ available to Napp is more direct and more
immediate than it is in other cases of predatory pricing.
Stated algebraically, the Tribunal found that the term (𝑝𝑝𝐵𝐵 − 𝑐𝑐𝐵𝐵 )𝜕𝜕𝑥𝑥𝐵𝐵 , particularly the size
𝜕𝜕𝑥𝑥𝐴𝐴
of the margin (𝑝𝑝𝐵𝐵 − 𝑐𝑐𝐵𝐵 ), represented successful recoupment and could not be used to
justify the fact that (𝑝𝑝𝐴𝐴 − 𝑐𝑐𝐴𝐴 ) < 0.
Some readers might assert that Napp is not a platform because it does not facilitate
interactions between the two sides. But whatever label one attaches to it, the logical
structure of the analysis is identical to that of a two-sided market. Moreover, this type of
effect could arise in settings that are widely agreed to constitute multi-sided markets
when platforms have sufficiently different characteristics from one another that the price
structure affects the ability of some firms to compete. For example, competing media
platforms may have very different business models (e.g. subscriber versus advertising-
supported business models), and a dominant firm might deviate from its otherwise
optimal business model (say by giving away subscriptions rather than charging for them)
precisely to harm rival platforms relying on different business models. 32
The logic of the no-economic-sense test does address such differences, and it indicates
that a pair of different questions should be posed with respect to recoupment: Is below-
cost pricing profitable for the platform because it makes the platform a stronger
competitor by building up its user base? Or is the below-cost pricing profitable only
because it also weakens competition by preventing rivals from building their own user
bases?
In answering these questions, it is important to recognise that, in the presence of network
effects, exclusionary behaviour can significantly harm competition and consumer welfare
without driving competitors from the market. As illustrated by in the Napp case,
weakening a rival can allow a dominant firm to charge higher prices and earn greater
profits even if the rival is not driven from the market entirely. 33 One might argue that
having a bright-line test based on exit would be useful because it provides greater
certainty and is easier to apply. However, the use of a bright line also raises the
possibility of gaming: a platform engaging in exclusion may seek to weaken its rivals just
up to the point that they are about to exit, while rivals might exit in order to trigger the
possibility of receiving damages that would be unavailable to them if they remained in
business.
It can be very challenging to determine whether below-cost pricing is profitable only
because it also weakens competition by preventing rivals from building their own user
bases. To do so, one might have to determine whether the firm’s conduct would be
profitable in a counterfactual world in which competitors were not weakened (i.e. that
rivals could continue to offer the same surplus that they otherwise would have). In the
presence of inter-temporal network effects, it becomes necessary to project the future
industry equilibrium in order to apply the test. Doing so can be very difficult given role of
consumer expectations and potentially complex business strategies.
Economists frequently assert that effects -rather than intentions- are what matter for
welfare and, thus, intentions are irrelevant. However, if one expects business people to
know what they are doing, then their views (expressed in ordinary-course-of-business
documents) may shed light on facts that are otherwise hard for an outsider to observe (e.g.
whether particular conduct made economic sense for non-exclusionary reasons). Of
course, there are issues relevant for competition policy that executives may be unqualified
to analyse, and there is a risk that companies will create documents solely with potential
litigation in mind. Hence, evidence of intention alone is insufficient to establish
anticompetitive effect or its absence. But neither is such evidence entirely uninformative.
As a general matter, it may be easier to determine when to find that a firm is not liable.
For example, it may be possible to rule out predatory pricing when it is clear that there
could have been little prospect of significantly weakening rivals (e.g. when rivals have
ready access to capital, the costs of multi-homing are low, and users are not locked-in to a
platform as the result of platform-specific investments or the inability of users to
co-ordinate on switching to another platform if it would benefit them collectively).
It is useful to discuss these issues in the context of an example. The following discussion
takes at face value certain claims made by the Initiative for a Competitive Online
Marketplace (ICOMP), an organisation funded by Microsoft. 34 According to ICOMP,
Google France neither charged map users (either consumers or the users of Google’s map
API) nor sold advertising. 35 Hence, at least according to ICOMP, the issue was not that
Google was pursuing a misunderstood two-sided market strategy. Instead, Google was
allegedly engaged in predation whereby, in the short run, it charged zero prices to both
sides of the platform and, in the long run, it would raise prices to both sides once it had
This section examines the treatment of exclusivity strategies with which a platform with
substantial market power seeks to weaken competition by demanding that some or all
user groups refrain from patronising competing platforms.
There are several different means by which a platform might limit multi-homing. The
most direct means is the imposition of contractual terms that prohibit a user from
participating on a platform if the user participates on any competing platform. 40
Exclusivity can also be indirectly induced by utilising price structures that make it
economically unattractive for a platform user to multi-home. Examples include quantity
discounts (such as volume-insensitive, or lump-sum, charges for platform use), as well as
discounts based on the percentage of a users’ patronage that is over a given platform (so-
called loyalty discounts). 41 Technological choices can also influence the cost of multi-
homing. For instance, a video game console manufacturer might adopt technical
standards that make it more costly for games created for one console to be ported to a
different brand of console. 42 Lastly, Hermalin and Katz (2006) show that the equilibrium
extent of multi-homing can be affected by the allocation of the authority to choose the
platform over which a transaction takes place when a pair of users on the two sides of the
market are on multiple platforms in common. Specifically, the side that lacks formal
authority may single home in order to force the hand of the side with formal authority.
As Lee (2013) points out, in order to understand the effects of imposing multi-homing
restrictions on users on one side of a platform, it is necessary to account for the reactions
of users on the other side of the platform. A reduction in multi-homing by users on one
side might lead to greater multi-homing by users on another. For example, if video game
developers are blocked from simultaneously offering their games on multiple brands of
video game console, then some gamers may respond by purchasing multiple brands of
consoles.
Before considering the effects of platform exclusivity, several leading theories of harm to
competition that have been developed in non-platform settings are reviewed. The
application of these theories to platforms and the implications for policy enforcement are
then discussed.
accept a proposal that harms dealers collectively even if that dealer receives very little
compensation for doing so. 47,48 As long as no one buyer is large enough to allow an
entrant to achieve a viable scale, a similar pattern can hold with respect to buyers
agreeing to exclusive relationships with an incumbent seller.
The second way Bork’s pay-for-exclusivity argument breaks down is more subtle.
Calzolari and Denicolò (2015) demonstrate that, even without contractual externalities, it
may be possible to attain exclusivity at no cost. Specifically, they analyse the
consequences of the fact that sellers typically face heterogeneous buyers and are unable to
engage in perfect price discrimination, so that, even under monopoly pricing, all but the
marginal buyers typically earn strictly positive surplus, or information rents. Calzolari
and Denicolò (2015, p. 3332) show that there is a sense in which this surplus can be used
as payment to buyers for agreeing to be exclusive. Because buyers would have received
this surplus in the form of information rents absent exclusivity, the exclusivity is
purchased by the seller at no cost. 49
Of course, the absence of a general proof that exclusive dealing is efficient does not prove
that exclusive dealing harms competition. There are, however, several theories under
which exclusive dealing can harm competition and consumers. All of these theories rely
on the existence of some asymmetry among manufacturers, but the nature of those
asymmetries, and the mechanisms by which competition is harmed, are very different.
The first two theories of harm are based on the assumptions that a manufacturer’s profits
are an increasing function of its rivals’ costs and that exclusivity arrangements can serve
as a means of raising those costs. The core difference between these two theories is
source of asymmetry among firms and the role of long-term contracts. The first theory
relies on temporal asymmetries. Specifically, it applies to situations in which an
incumbent supplier can “tie up” dealers or other trading partners (e.g. buyers) before a
competing supplier is able to enter the market. The supplier induces the other parties to
agree to long-term, exclusive contracts such that, if a competing supplier later entered the
market, it would be unable to trade with the parties under contract. If the contracts have
staggered expiration/renewal dates, then there will be no date on which an entrant could
freely compete for all potential trading partners. In the presence of economies of scale,
the entrant’s resulting level of activity might be too small to be economically viable even
if some trading partners remained available.50
The second theory of harm is also based on the assumption that a manufacturer benefits
from increases in its rivals’ costs. If there are economies of scale and scope in
distribution, then a system of exclusive dealers will raise the distribution costs of all
manufacturers but will do so more for smaller ones than larger ones. The net effect may
be to raise the profits of the largest manufacturer, even though its costs of distribution are
raised. 51 Hence, if there is some source of asymmetry that results in one manufacturer’s
having much larger sales than others, then that manufacturer can have incentives to seek
exclusivity. Notice that contracts do not play a commitment role under this theory -
dealers can be free to switch to other manufacturers. The relevant asymmetry is with
regard to the manufacturers’ sizes (and thus their abilities to generate sales to support
exclusive dealer networks) rather than the order in which they enter into contractual
negotiations with dealers.
The third theory is not based on raising rivals’ costs through the denial of scale. Calzolari
and Denicolò (2015) examine competition between duopolists offering differentiated
products, where one of the firms -the “dominant” supplier- has a cost or vertical quality
advantage. Because products are differentiated and buyers have a taste for variety, the
higher-cost firm can still compete for sales at the margin if buyers are able to patronise
both sellers simultaneously. As Calzolari and Denicolò (2015, p. 3322) explain, this fact
can create an incentive to impose exclusivity 52:
[If] exclusive contracts are banned, firms are forced to compete for each
marginal unit of a buyer’s demand. Excluding rivals thus requires a limit pricing
strategy, which in turn entails a sacrifice of profits. When exclusive contracts are
permitted, on the other hand, firms compete for the entire volume demanded by a
buyer—i.e., competition is in “utility space.” In utility space, the dominant firm
can exclude rivals by leveraging on the information rents left on inframarginal
units. If the competitive advantage is large, the dominant firm can keep charging
monopoly prices and exclude rivals by means of exclusivity clauses only. If the
competitive advantage is more limited, exclusive prices cannot be set at the
monopoly level, but the discount required to foreclose is smaller than it would be
in the absence of exclusive contracts. [Emphasis added, internal footnote
omitted.]
It is widely recognised that, in addition to harming competition, exclusive arrangements
can also create a new dimension of competition: competition for exclusivity. Moreover,
Calzolari and Denicolò identify a specific mechanism through which exclusivity can
strengthen, rather than weaken completion overall. As Calzolari and Denicolò (2015, p.
3323) explain:
Whereas product differentiation softens competition for marginal units, it does
not soften competition in utility space. In utility space, product diversity is in fact
irrelevant: all that counts is the amount of rent left to buyers. When firms [have
comparable cost or vertical quality levels], this tends to make competition in
utility space tougher than competition for marginal units.
Thus, when the suppliers are differentiated but have relatively similar costs or (vertical)
quality levels, the effect of exclusivity can be to intensify competition by switching it
from differentiated competition for marginal units to undifferentiated competition in
utility space.
Applicability to platforms
Several features of platform markets make them susceptible to the use of exclusive
agreements to harm competition. First, the cross-platform nature of network effects gives
rise to the possibility of contractual externalities when there is no mechanism for users on
one side of a platform to make financial transfers to users on the other side in order to
influence their choice of platform. Absent such mechanisms, a user on one side of a
platform might have little concern for the effects of a decision to single-home on the
welfare of users on another side of the platform. 53
Second, cross-platform network effects give rise to demand-side economies of scale that
allow a platform to benefit if it can use exclusivity as a means of limiting participation on
rival platforms and, thus, raising rivals’ costs (i.e. weakening their ability to provide user
benefits). Moreover, the provision of multi-sided platform services may be subject to
strong production economies of scale in addition to demand-side economies of scale,
reinforcing these effects. Hence, if there is some initial asymmetry, the leading or
dominant platform may be able to benefit from imposing conditions that drive most users
in one or more groups to single-home on that platform when they would otherwise have
multi-homed. The dominant platform can benefit from increases in its rivals’ average
costs if the higher costs drive the rivals from the market. And -because the rivals will be
weaker competitors- the dominant platform can benefit from increases in its rivals’
marginal costs of generating user benefits even if the rivals remain in the market.
Shapiro (1999, p. 677) presents a dynamic theory of these effects and argues that multi-
homing can serve as a transitional user strategy that facilitates entry by new platforms.
The logic of this argument is that, faced with an all-or-nothing choice between an
emerging platform and an established one, there are conditions under which users will
choose the established platform. However, given the option of multi-homing, some
consumers might do so, allowing the emerging platform to begin to build an installed
base that will then attract further users. By imposing an exclusivity requirement, an
incumbent platform can eliminate this path to entry. Shapiro (1999, pp. 680 and 683) also
argues that exclusivity can lead to pessimistic consumer expectations regarding the
entrant’s prospects, which reinforce this effect. 54
Turning to the sources of asymmetries, platforms may have different production costs,
product attributes, or market entry dates. As a result of these differences, platforms may
differ in terms of their existing installed bases and/or users’ expectations regarding the
number of users who will patronise the platforms in the future. There can also be
feedback effects that reinforce initial asymmetries (e.g. if there are expected to be more
side-A users on a platform, then more side-B users are attracted, which then leads more
side-A users to patronise the platform and starts a new round of feedback). 55
Four legal cases illustrate how the courts have treated platform exclusivity agreements.
These cases also demonstrate that the issues are not new.
The earliest of these cases involved media platforms. The Lorain Journal was the only
daily newspaper in Lorain, Ohio. 56 In 1948, the radio station WEOL began broadcasting
in an area that included the Journal’s subscribers. The Journal demanded that advertisers
single-home (i.e. it refused to sell advertising to any business that purchased advertising
from WEOL). The U.S. Department of Justice alleged (and the Supreme Court agreed)
that this conduct was exclusionary and intended to harm competition by driving WEOL
out of the local market for advertising. This case fits Calzolari and Denicolò’s (2015)
theory. The key asymmetry was that, because of the nature of radio advertising and the
fact that Journal had a much larger audience than did WEOL, advertisers wanted to use
advertising on WEOL only as a supplement to advertising in the Journal. 57 Calzolari and
Denicolò’s theory indicates that the Lorain Journal was able to weaken competition,
which resulted in greater unit sales of advertising by the Journal and higher advertising
prices, to advertisers’ detriment.
The next two cases also build on the idea that, if faced with an all-or-nothing choice,
users will choose to patronise the platform with the largest user base but otherwise would
multi-home. One case involved floral delivery platforms, which create value by bringing
together florists receiving orders for flowers with florists fulfilling orders. Specifically, if
a consumer desires to send flowers to someone in another city, the consumer can place an
order with a local florist that is a member of a floral-delivery platform and that order will
be fulfilled by another platform member that is located near the recipient of the flowers.
In the mid-1950s, FTD was by far the largest such platform in the United States and had a
policy directly prohibiting its member florists from participating in competing floral
platforms. 58 In 1956, the U.S. Department of Justice filed a complaint against FTD
alleging that its exclusive membership restriction eliminated competition and preserved
FTD’s market dominance. FTD and the Department entered into a consent decree
enjoining conduct that had the purpose or effect of imposing exclusivity. 59 In 1995, the
Department alleged that FTD had violated the consent decree by offering financial
rewards to florists that were members of only FTD and that FTD did so in order to
weaken rival platforms’ ability to compete.60 FTD agreed to cease offering the rewards. 61
The case thus illustrates an asymmetry based on florists’ expectations of platform size
and the use of both direct and indirect measures to induce single-homing.
In the late 1980s, the then-leading video game console manufacturer Nintendo used a
direct measure by requiring companies developing games for its Nintendo Entertainment
System console to release those games exclusively on that platform for a period of two
years. Rival console maker Atari sued Nintendo, alleging that this practice harmed
competition and preserved its market position. 62 Although Atari lost the case, Nintendo
ceased the practice before the verdict was reached. 63
Lastly, in 2001, the U.S. Department of Justice successfully argued that the MasterCard
and Visa credit card networks harmed competition by prohibiting certain forms of multi-
homing. 64 MasterCard and Visa both had policies that limited member banks’ abilities to
issue cards on competing credit and charge card platforms, namely American Express and
Discover/NOVUS. Because of asymmetries in coverage and the card products supported
by the platforms, banks were reluctant to forego card issuing on MasterCard and Visa
entirely in order to issue credit and charge cards on American Express and/or
Discover/NOVUS. However, there was evidence that some banks issuing cards on
MasterCard and Visa would be interested is issuing cards on the American Express or
Discover/NOVUS networks if multi-homing were permitted. 65 American Express’s and
Discover/NOVUS’s inability to attract these card-issuing banks weakened platform
competition because the two platforms were less attractive both to cardholders and -on
the other side of the platforms- merchants. After the rules were dropped, several banks
began issuing cards on the American Express and Discover platforms.
In several respects, this case, too, is a good match for Calzolari and Denicolò’s (2015)
theory. At the time, both American Express and Discover/NOVUS were seen more as
niche networks (with American Express supporting cards aimed at high-end consumers
and Discover supporting cards aimed at low-end consumers), while MasterCard and Visa
supported cards aimed at a broad range of consumers. Hence, American Express and
Discover were better positioned to compete for marginal business than compete in utility
space.
from the market. Moreover, the mechanism of harm identified by Calzolari and Denicolò
(2015) relies on shifting the nature of competition rather than eliminating competitors. 67
Third, enforcers should be careful not to place undue weight on contract length.
Contractual lock-in is important under a theory of harm in which the asymmetry
facilitating the use of exclusivity is temporal and the incumbent uses long-term, staggered
contracts signed before the entrant is present to make entry more costly. However, the
other theories of harm discussed above do not rely on contracts as commitments and,
thus, contract length is unimportant. Instead there has to be an asymmetry among
suppliers in terms of costs, product quality, user bases, or user expectations. In the U.S at
least, courts have moved away from reliance on contract length. For example, the
Dentsply appellate court focused on “the nature of the relevant market and the established
effectiveness of the restraint” rather than contract length. 68
Fourth, Calzolari and Denicolò (2015, 3345-46) find that, when exclusivity shifts the
market from competition for marginal units to competition for a user’s entire volume, it
can strengthen or weaken competition, depending on the degree of asymmetry between
different suppliers. The authors also indicate that exclusives are less likely to harm
competition when rivals also impose exclusivity (Id.). This logic suggests that, when
platforms are similar and all impose exclusivity, they are doing so for reasons other than
harming competition by weakening some firms’ abilities to compete relative to others’.69
Fifth, as discussed above for non-platform markets, exclusivity can create new avenues of
competition (e.g. competition for exclusivity), which complicates enforcement. This is
also true of platform markets. When users on one side single home and users on the other
do not, the single-homing side chooses the platforms over which interactions will occur.
Hence, platforms engage in price competition to attract users on the single-homing side,
but not users on the multi-homing side. Indeed, each platform has a monopoly for access
to its single-homing users. 70 By contrast, when multi-homing is blocked, platforms will
compete for users on both sides.
Building on this observation, Armstrong and Wright (2007) show that exclusivity
requirements can have very strong implications for the distribution of economic surplus
between two sides of platform users. Inter alia, Armstrong and Wright analyse a model of
competition between two platforms that facilitate interactions between buyers and sellers
in which the authors reach the following findings. Platforms compete solely for buyers
(who single-home) and extract all of the surplus from sellers (who multi-home) when
platforms cannot require sellers to be exclusive. By contrast, platforms compete to attract
sellers to exclusive relationships and extract all of the surplus from buyers when
exclusive contracts are permitted.
The possibility of such dramatic differences in the effects of exclusivity on the welfare of
different user groups raises an important question for competition policy. How should the
shift in surplus be treated? One view is that a user-welfare standard should weigh all users
equally and focus solely on the net effects. An alternative view is that each user group is
entitled to the benefits of competition and that harm to one user group due to harm to
competition cannot be offset by gains to another user group that are a consequence of the
loss of competition. It would be useful to have greater clarity regarding policy objectives.
Although the case did not involve exclusivity, recent litigation between the U.S.
Department of Justice and American Express has brought this issue to the fore. 71 The
Department of Justice argued that American Express’s conduct harmed competition in the
market for credit and charge card acceptance services sold to merchants and that
demonstrating harm to merchants was sufficient to shift the burden to American Express
to show that it had an offsetting, pro-competitive rationale. Although the Department of
Justice prevailed at trial, the appellate court overturned on the grounds that the
government should have proven that the losses suffered by merchants as a result of
American Express’s conduct were not outweighed by gains to American Express’s card
holders. 72
Sixth, as is well known, network effects can give rise to natural monopoly conditions. A
potentially challenging question for competition policy enforcers is whether the greater
realisation of network effects due to the elimination of rival networks and the consequent
coalescing of all users on the same network could be considered to be an efficiencies
defence. For example, Armstrong and Wright (2007) consider a model of competition
between undifferentiated platforms and find that exclusivity can be used to eliminate a
rival. However, exclusion is efficient in their model. More generally, exclusion could also
occur with a small degree of product differentiation, in which case, it could be inefficient
if the loss of differentiation benefits exceeded the costs of multi-homing (which -absent
direct or indirect restraints- could serve as a means of fully realising cross-platform
network effects).
Seventh, traditional types of efficiencies should also be credited, where valid. For
example, translating the leading pro-competitive justification for exclusivity to platforms,
a platform might argue that exclusivity increases its willingness to make investments that
benefit users. Segal and Whinston (2000) find that exclusivity has this effect only if the
platform’s investments raise users’ value of transacting with rival platforms, so that a
commitment from the user is needed to prevent free riding. Thus, a platform would have
to demonstrate that it is investing in its users in ways that raise the value those users
would generate if they were to patronise a rival platform.
Lastly, the lack of an efficiency rationale justifying the imposition of exclusivity can be
informative. In the Dentsply, Lorain Journal, and Visa cases, for example, the defendants
were unable to produce credible efficiency rationales for their challenged conduct.
5. Conclusion
Appendix
The following highly stylised model illustrates why what is ostensibly a total-surplus
standard could, in practice, become a competitor-surplus standard. Suppose the actual
change in total surplus due to certain conduct is: ∆𝑊𝑊 = ∆𝜋𝜋 𝐼𝐼 + ∆𝜋𝜋 𝑅𝑅 + ∆𝑆𝑆, where the three
components of the change in welfare are the change in the incumbent’s profits, the
change in a rival’s profits, and the change in consumer surplus, respectively. In addition,
suppose both the decision-making firm and its rival know all of the values with certainty
but the court observes only ∆𝑊𝑊 + 𝜀𝜀, where 𝜀𝜀 is a random observation error. Let
𝜌𝜌(∆𝑊𝑊) ≡ 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃{∆𝑊𝑊 + 𝜀𝜀 < 0} denote the probability that firm will be found liable.
Now, consider the rival’s incentive to initiate an enforcement action. Suppose that,
conditional on the defendant’s being found guilty, the expected change in the rival’s
profits is equal to 𝛾𝛾∆𝜋𝜋 𝑅𝑅 , where 𝛾𝛾 > 0 is a factor that accounts for both the expected
amount of monetary damages awarded by the court for past harm and the net present
value of not being subject to the defendant’s adverse conduct in the future. Ignoring any
litigation costs, the expected change in the rival’s profits is 𝜌𝜌(∆𝑊𝑊)𝛾𝛾∆𝜋𝜋 𝑅𝑅 .
The rival will bring a complaint if and only the expected benefits are greater than the
costs. Letting L denote the rival’s cost of litigation, it will bring a complaint if and only if
𝜌𝜌(∆𝑊𝑊)𝛾𝛾∆𝜋𝜋 𝑅𝑅 > 𝐿𝐿. This rule is equivalent to bringing a complaint if and only if ∆𝜋𝜋 𝑅𝑅 < 0
and ∆𝑊𝑊 < 𝛿𝛿, for some constant 𝛿𝛿 which may be greater or less than 0, depending on the
values of 𝛾𝛾 and L and the distribution of 𝜀𝜀.
Given this decision rule, the potential defendant has incentives to avoid actions that harm
its rivals -whether through exclusion or competition on the merits. Such an implicit rule is
not entirely bad if there is a positive correlation between the amount of harm to the rival
and the amount of harm to consumers. However, the correlation might well be negative
because stronger competition by one supplier typically will benefit consumers but lower
the profits of a rival supplier. Indeed, the sign of the correlation might be seen as a
measure of whether the harm to rival is the result of competition or exclusion.
Notes
1 This is similar to the definition suggested by Weyl (2010). For a discussion of issues concerning
the definition of multi-sided platforms and markets, see Hermalin and Katz (forthcoming).
2 Given space constraints, this paper does not address issues of market definition and the
assessment of market power, which are often critical in litigation and the determination of whether
the defendant’s conduct can cause material harm. These issues are addressed by other
contributions to this workshop. That said, there is a risk of error when making this separation
because the various issues interact with one another and should be addressed in an integrated
analysis. For example, market definition is not an end in itself, and it should be closely tied to the
specific questions at hand with respect to the conduct at issue.
3 The distinction between the two concepts is not an entirely sharp one. For example, in markets
with learning-by-doing, predatory pricing may raise rivals’ costs by denying them sales that would
have otherwise led to learning and lower costs. Similarly, in markets with network effects,
predatory pricing can result in rival platforms’ having fewer users and—because network benefits
are reduced—higher costs of providing any given level of user benefits.
4 Various forms of this test have been advocated by Steven Salop. (See, e.g., Salop (2006).)
5 For a discussion of other problems, see Melamed (2005) and references therein.
6 For example, the U.S. Supreme Court has stated that: “The mere possession of monopoly power,
and the concomitant charging of monopoly prices, is not only not unlawful; it is an important
element of the free-market system. The opportunity to charge monopoly prices—at least for a
short period—is what attracts “business acumen” in the first place; it induces risk taking that
produces innovation and economic growth. To safeguard the incentive to innovate, the possession
of monopoly power will not be found unlawful unless it is accompanied by an element of
anticompetitive conduct.”[Emphasis in original.] (Verizon Communications Inc. v. Law Offices of
Curtis V. Trinko, LLP, 540 U.S. 398, 407 (2004).)
7 See Appendix.
8 This test is generally associated with Judge Richard Posner. (See, e.g., Posner, 2001, pp. 94–95.)
For a particular application of this logic to develop a cost test for predatory pricing, see Baumol
(1996).
9 European Commission (2009), ¶¶ 23 and 67. In assessing predatory pricing, the Commission also
examines whether the alleged predator is engaged in short-run profit sacrifice, a variant of the next
standard for exclusionary behaviour discussed below. (Id., ¶63.)
10 Id., ¶ 24.
11 Id.
12 See, e.g., Werden (2006). Ordover, J. A., and Willig, R. D. (1981) advocate a definition of
predatory behaviour along similar lines but focused on exit rather than considering all degrees of
harm to competition.
13 See, e.g., Brief of the Appellees United States and the State Plaintiffs at 48, United States v.
Microsoft Corp., 253 F.3d 34 (D.C. Cir. 2001) (Nos. 00-5212, 00-5213); Brief for Appellant
United States at 2, 30, United States v. AMR Corp., 335 F.3d 1109 (10th Cir. 2003) (No. 01-3202)
(public redacted version); Brief for Appellant United States at 28, United States v. Dentsply Int’l,
Inc., 399 F.3d 181 (3d Cir. 2005) (No. 03-4097) (public redacted version).
14 See, e.g., Melamed (2006).
15 Brooke Group, Ltd. v. Brown & Williamson Tobacco Corporation, 509 U.S. 209, 222 (1993).
Many other national competition authorities consider price-cost tests as well. For a survey of
practices, see Unilateral Conduct Working Group (2008), § II.1.
16 Id. at 224.
17 Case C-62/86 AKZO Chemie v Commission, [1991] ECR I-3359; Case C-333/94 P Tetra Pak
International v. Commission [1996] ECR I-5951. The cost concepts are stated for a single-product
firm. For the corresponding multi-product concepts, see European Commission (2009), ¶ 26.
18 Case C-333/94 P Tetra Pak International v Commission, [1996] ECR I-5951; Case C-202/07 P
France Télécom SA v Commission of the European Communities, [2009] ECJ First Chamber.
19 Unilateral Conduct Working Group (2008), § II.2.A and B. The treatment of recoupment varies
among national competition authorities within the European Union, as well as among the
authorities of other nations. For a survey of individual countries’ practices, see id., § II.2.
20 European Commission (2009), ¶¶ 20 and 68-71.
21 Areeda and Turner (1975).
22 For a discussion of the debate, see Elhauge (2003).
23 Although many commentators prefer to focus on two-sided platform issues, the use of the
freemium model appears to have played a central role in a recent case brought against Google
regarding the pricing of its mapping application. (Bottin Cartographes v. Google France, Cour
d’appel, Paris Pôle 5, Chamber 4, 25 November 2015.)
24 For an early formal model of network competition with below-cost pricing, see Katz and Shapiro
(1986).
25 Similar issues arise with learning by doing. See Cabral and Riordan (1994 and 1997).
26 Vasconcelos (2015) extends some of Farrell and Katz’s (2005) results to platforms with one- and
two-sided cross-platform network effects.
27 See, e.g., Wright (2004).
28 Algebraically, the change in 𝑥𝑥𝐵𝐵𝑖𝑖 equals (𝜕𝜕𝑥𝑥𝐵𝐵𝑖𝑖 ⁄𝜕𝜕𝑥𝑥𝐴𝐴𝑖𝑖 )(𝑑𝑑𝑥𝑥𝐴𝐴𝑖𝑖 ⁄𝑑𝑑𝑝𝑝𝐴𝐴𝑖𝑖 ) + (𝜕𝜕𝑥𝑥𝐵𝐵𝑖𝑖 ⁄𝜕𝜕𝑥𝑥𝐴𝐴−𝑖𝑖 )(𝑑𝑑𝑥𝑥𝐴𝐴−𝑖𝑖 ⁄𝑑𝑑𝑝𝑝𝐴𝐴𝑖𝑖 ). The
incorrect approach described in the text amounts to assuming that the observed change in 𝑥𝑥𝐵𝐵𝑖𝑖
equals (𝜕𝜕𝑥𝑥𝐵𝐵𝑖𝑖 ⁄𝜕𝜕𝑥𝑥𝐴𝐴𝑖𝑖 )(𝑑𝑑𝑥𝑥𝐴𝐴𝑖𝑖 ⁄𝑑𝑑𝑝𝑝𝐴𝐴𝑖𝑖 ). This approach would thus credit the effects of weakening the rival
as benefits realised due to competition on the merits.
29 This summary of this matter is based on Case 1001/1/1/01 Napp Pharmaceutical Holdings Limited
and Subsidiaries v Director General of Fair Trading [2002] CAT 1.
30 Id., ¶ 51.
31 Id., ¶ 261.
32 Hoernig (2007) examines the potential for a mobile telecommunications provider to use a high
differential between off-net and on-net prices to harm competition among asymmetric providers.
33 See, also, European Commission (2009), ¶ 69.
34 No representations are being made here regarding the veracity of ICOMP’s factual claims.
35 Initiative for a Competitive Online Marketplace (Undated), p. 10.
36 There also was an issue whether the free version served as a promotional tool to induce map users
to purchase a paid service. This is not a two-sided issue.
37 Bottin Cartographes v. Google France, Cour d’appel, Paris Pôle 5, Chamber 4, 25 November
2015, p. 8.
38 Id., pp. 8-9. Recall that, as a general matter, exclusionary conduct that weakens but does not
eliminate rivals can also harm competition and consumers.
39 Using the notation introduced earlier in the text, these commentators would assert that it would
sufficient to compare 𝑝𝑝𝐴𝐴 + 𝑝𝑝𝐵𝐵 with 𝑐𝑐𝐴𝐴 + 𝑐𝑐𝐵𝐵 to answer any relevant questions regarding market
power, profitability, or harm to competition, without regard to the values of the individual prices
the make up the sum.
40 Lee (2013) notes that, in some industries, a platform can impose exclusivity by vertically
integrating into one side of the market. For example, video game consoles are platforms that bring
together game developers and gamers. An established strategy for console manufacturers is to
integrate into the development of games that are offered exclusively on their platforms.
41 For an analysis of loyalty discounts in “traditional” markets, see Calzolari and Denicolò (2013)
and Klein and Lerner (2016). The earlier paper demonstrates that, under some conditions, market-
share discounts and exclusivity requirements can have very different competitive effects from one
another.
42 In many settings, blocking multi-homing is not the same as blocking platform compatibility. With
compatible platforms, there is a single “network,” and users on side A of one platform benefit
from actions taken by another platform to increase the number of users on its B side. Such effects
need not arise with incompatible platforms even when there is no prohibition on multi-homing and
the act of multi-homing is costless. That said, there are similarities in that a dominant network can
have incentives to oppose compatibility in order to weaken rivals. For an insightful early analysis,
see Cremer et al. (2000).
43 Framed in terms of platforms, one could argue that a manufacturer is a platform that facilitates
transactions between dealers and consumers. This view is counterintuitive but it speaks to the lack
of agreement regarding what constitutes a platform. It is perhaps more intuitive to think of dealers
as platforms and a manufacturer as a platform user which is demanding that other potential users
be excluded.
44 Brief for the United States, Redacted -- Public Version, United States v. Dentsply International,
Inc., p. 18.
45 Bork (1978, p. 309). For an excellent survey of the literature responding to this argument, see Rey
and Tirole (2007, § 4).
46 Aghion and Bolton (1987) demonstrate that, by signing a long-term exclusive dealing agreement
with penalty clauses, a dealer and incumbent manufacturer can force a manufacturer that later
enters the market to compensate the dealer for breaking the long-term agreement. In this way, the
dealer and incumbent manufacturer can appropriate some of the benefits of entry for themselves,
which can reduce entry and harm consumers as well as the potential entrant.
47 This common intuition is summarised in Katz (1989, p. 708). It is formalised and more fully
explored by Rasmussen et al. (1991). Segal and Whinston (2000) correct certain inconsistencies in
Rasmussen et al.’s analysis.
48 Strikingly, as pointed out by Rasmussen et al. (1991, p. 1143), a manufacturer need not have
market power prior to the imposition of exclusive dealing in order to be able profitably to sign
most or all dealers to exclusive contracts. This finding indicates that, in applying a market-power
threshold to screen potential cases, competition policy enforcers should assess what the degree of
market power is when exclusivity is in place.
49 This form of exclusion need not entail any sacrifice of profits. Hence, tests based on evidence of
profit sacrifice could fail to detect this type of harm to competition.
50 In this form of the theory, exclusivity is used to deter entry. As noted above, Aghion and Bolton
(1987) show that exclusive contracts can also be used to extract rents from entrants rather than
deter entry entirely.
51 Katz and Rosen (1985) and Seade (1985) showed that when marginal costs are increased by some
action, even a symmetric (across all manufacturers) cost increase may raise a manufacturer's
profits.
52 In Calzolari and Denicolò’s model, the dominant firm offers both exclusive and non-exclusive
pricing terms. However, their model can be extended to situations in which the seller must choose
one form of pricing or the other.
53 As Rochet and Tirole (2006 , p. 646) observed, this is an important difference between multi-sided
platforms and the sale of complementary goods: unlike the case of buyers who purchase
complementary goods (or “systems”), the decision makers on the different sides of a platform
generally are not concerned with one another’s welfare.
54 Shapiro (1999) provides a verbal analysis for generic network effects. Doganoglu and Wright
(2010) further explore the effects of exclusivity on entry in a formal model that explicitly
examines two-sided markets.
55 These asymmetries need not always favor the incumbent. Shapiro (1999, p. 682) observes that an
entrant with a sufficiently superior technology might wish to impose exclusivity to hasten users’
switching to it. In other words, a technological asymmetry favoring the entrant might outweigh a
temporal asymmetry favoring the incumbent.
56 This case description is based on Lorain Journal Co. v. United States, 342 U.S. 143 (1951).
57 For a range of alternative interpretations, see Lopatka and Kleit (1995).
58 This case description is based on United States v. FTD Corp.; Florists’ Transworld Delivery, Inc.;
FTD Ass’n, Supplemental to Civil Action No. 56-15748, Memorandum of the United States in
Support of Proposed Enforcement Order, July 1, 1995.
59 United States v. Florist’s Telegraph Delivery Ass’n, 1956 Trade Cas. (CCH) ¶ 68,367 (E,D, Mich.
1956).
60 United States v. FTD Corp., 60 Fed. Re. 40,859 (E.D. Mich. 1995).
61 Shapiro (1999, p. 677).
62 Atari Corp. V Nintendo, No. 89-0824 (N.D.Cal. 1992); Atari Games Corp. and Tengen, Inc. v.
Nintendo of America Inc. and Nintendo Co., Ltd., 975 F.2d 832 (Fed. Cir. 1992).
63 Jonathan Weber, “Jury Sides With Nintendo in Suit Brought by Atari: * Verdict: Panel exonerates
the video game giant of trying to monopolise the market and rules it had not damaged its rival.”
Los Angeles Times, May 2, 1992
64 United States v. Visa U.S.A., Inc., 163 F. Supp. 2d 322, 338 (S.D.N.Y. 2001); United
States v. Visa U.S.A., Inc., 344 F.3d 229, 238-39 (2d Cir. 2003). I was retained by the
U.S. Department of Justice as an economic expert in this matter, and I testified at trial.
65 There was also a horizontal element of harm. At the time, MasterCard and Visa were associations
collectively governed by member card-issuing banks, and the policies at issue restricted the means
by which they could compete with one another (i.e. it limited their choices of payment networks
on which to rely).
66 The appellate court in United States v. Microsoft Corp. addressed this issue, noting that U.S.
courts have “taken care to identify the share of the market foreclosed” (United States v. Microsoft
Corp., 253 F.3d 34 (D.C. Cir. 2001) p. 69). While observing that “[b]ecause an exclusive deal
affecting a small fraction of a market clearly cannot have the requisite harmful effect upon
competition, the requirement of a significant degree of foreclosure serves a useful screening
function…” (id.), the court went on to state that, “[a]t the same time, however, we agree with
plaintiffs that a monopolist's use of exclusive contracts, in certain circumstances, may give rise to
a § 2 violation even though the contracts foreclose less than the roughly 40% or 50% share usually
required in order to establish a § 1 violation.” (Id., p. 70.). The economics of network effects
suggests that competition could be harmed by exclusivity involving very substantially less than 40
percent of users.
67 The possibility of using exclusivity to harm platform competition without inducing exit has been
recognised by the U.S. courts. For example, in United States v. Visa U.S.A., Inc., 344 F.3d 229
(2d Cir. 2003), cert. denied, 125 S. Ct. 45 (2004), the defendants were found liable even though
the targets of exclusion were not forced to exit the market.
68 United States v. Dentsply International, Inc., 399 F.3d 181, 184 (3d Cir. 2005).
69 Hermalin and Katz (2013) examine a theoretical model in which exclusivity creates differentiation
that relaxes price competition and benefits all suppliers. The authors find that the welfare effects
are ambiguous when one accounts for investment competition. Lee (2013) conducts an empirical
study of the video game industry and finds evidence that, in that industry for the period he
examined, limitations on multi-homing by video game developers benefited the smaller, entrant
platforms by allowing them to differentiate themselves from the incumbent with the larger
installed base.
70 In telecommunications, this effect is known as the terminating access monopoly problem. For an
early general analysis, see Armstrong (2006).
71 The description of this case is based on United States et al. v. American Express Company,
MasterCard International Inc, Visa, Inc., No. 15-1672 (2d Cir. 2016) reversing 88 F. Supp. 3d 143
(E.D.N.Y. 2015). I was retained in this matter by the U.S. Department of Justice to serve as an
economic expert, and I testified at trial. The U.S. Supreme Court has granted certiorari and will
review this case in its 2017 term.
72 In my opinion, the evidence demonstrated that the harm to merchants and their customers, in fact,
outweighed the gains to American Express card holders.
References
Aghion, P., and Bolton, P. (1987). Contracts as a Barrier to Entry. American Economic Review 77(3),
388-401.
Areeda, P., & Turner, D. F. (1975). Predatory pricing and related practices under Section 2 of the
Sherman Act. Harvard Law Review, 697-733.
Armstrong, M. (2006). Competition in two‐sided markets. The RAND Journal of Economics, 37(3), 668-
691.
Armstrong, M., and Wright, J. (2007). Two-sided markets, competitive bottlenecks and exclusive
contracts. Economic Theory, 32(2), 353-380.
Baumol, W.J. (1996). Predation and the Logic of the Average Variable Cost Test. Journal of Law &
Economics, 39(1), 49-72.
Behringer, S., and Filistrucchi, L. (2015). Areeda–Turner in two-sided markets. Review of Industrial
Organization, 46(3), 287-306.
Bork, Robert H. (1978). The Antitrust Paradox: A policy at war with itself. New York: Basic Books.
Cabral, L.M.B., and Riordan, M.H. (1994). The Learning Curve, Market Dominance, and Predatory
Pricing. Econometrica, 62, 1115–1140.
Cabral, L.M.B., and Riordan, M.H. (1997). The Learning Curve, Predation, Antitrust, and Welfare. The
Journal of Industrial Economics, XLV, 155–169.
Calzolari, G., and Denicolò, V. (2013). Competition with exclusive contracts and market-share discounts.
The American Economic Review, 103(6), 2384-2411.
Calzolari, G., and Denicolò, V. (2015). Exclusive contracts and market dominance. The American
Economic Review, 105(11), 3321-3351.
Chowdhury, S. M., and Martin, S. (2010). Exclusivity and exclusion on platform markets. Journal of
Economics, 1-24.
Cremer, J., Rey, P., and Tirole, J. (2000). Connectivity in the commercial Internet. The Journal of
Industrial Economics, 48(4), 433-472.
Doganoglu, T., and Wright, J. (2010). Exclusive dealing with network effects. International Journal of
Industrial Organization, 28(2), 145-154.
Elhauge, E. (2003). Why Above-Cost Price Cuts To Drive Out Entrants Are Not Predatory—and the
Implications for Defining Costs and Market Power. Yale Law Journal, 112, 681-827.
European Commission. (2009). Guidance on the Commission’s enforcement priorities in applying
Article 82 of the EC Treaty to abusive exclusionary conduct by dominant undertakings, 2009/C
45/02), Official Journal of the European Union, C 45/7.
Farrell, J., and Katz, M. L. (2005). Competition or predation? Consumer coordination, strategic pricing
and price floors in network markets. The Journal of Industrial Economics, 53(2), 203-231.
Hermalin, B. E., and Katz, M. L. (2006). Your network or mine? The economics of routing rules. Rand
Journal of Economics, 37(3), 692-719.
Hermalin, B. E., and Katz, M. L. (2013). Product Differentiation through Exclusivity: Is there a
One‐Market‐Power‐Rent Theorem? Journal of Economics & Management Strategy, 22(1), 1-27.
Hermalin, B. E., and Katz, M. L. (forthcoming). What’s So Special about Two-Sided Markets? In
Economic Theory and Public Policies: Joseph Stiglitz and the Teaching of Economics, New York:
Columbia University Press.
Hoernig, S. (2007). On-net and off-net pricing on asymmetric telecommunications networks. Information
Economics and Policy, 19(2), 171-188.
Initiative for a Competitive Online Marketplace. (Undated). How Google Monopolised Online Mapping
& Listings Services.
Katz, M.L., and Rosen, H.S. (1985). Tax Analysis in a oligopoly model. Public Finance Quarterly 13(1),
3-20.
Katz, M. L., and Shapiro, C. (1986). Technology adoption in the presence of network externalities.
Journal of Political Economy, 94(4), 822-841.
Klein, B., and Lerner, A.V. (2016). Price-Cost Tests in Antitrust Analysis of Single Product Loyalty
Contracts. Antitrust Law Journal, 80(3), 631-679.
Lee, R.S. (2013). Vertical Integration and Exclusivity in Platform and Two-Sided Markets. American
Economic Review, 103(7), 2960–3000.
Lopatka, J. E., and Kleit, A. N. (1995). Mystery of Lorain Journal and the Quest for Foreclosure in
Antitrust. Texas Law Review, 73(6), 1255-1306.
Melamed, A. D. (2005). Exclusionary Conduct under the Antitrust Laws: Balancing, Sacrifice, and
Refusals to Deal. Berkeley Tech. LJ, 20, 1247.
Melamed, A. D. (2006). Exclusive dealing agreements and other exclusionary conduct—Are there
unifying principles? Antitrust Law Journal, 73(2), 375-412.
Ordover, J. A., and Willig, R. D. (1981). An economic definition of predation: Pricing and product
innovation. The Yale Law Journal, 91(1), 8-53.
Posner, R.A. (2001). Antitrust Law, 2d ed. Chicago: University of Chicago Press.
Rasmusen, E.B., Ramseyer, J.M., and Wiley, Jr., J.S. (1991). Naked Exclusion. The American Economic
Review 81(5), 1137-1145.
Rey, P. and Tirole, J. (2007). A primer on foreclosure. In Mark Armstrong and R. Porter, editors.
Handbook of Industrial Organization, Volume 3. Elsevier B.V., 2145-220.
Rochet, J. C., and Tirole, J. (2006). Two‐sided markets: a progress report. The RAND Journal of
Economics, 37(3), 645-667.
Salop, S.C. (2006). Exclusionary Conduct, Effect on Consumers, and the Flawed Profit-Sacrifice
Standard. Antitrust Law Journal, 73, 329–33.
Seade, J. (1985). Profitable Cost Increases and the Shifting of Taxation : Equilibrium Response of
Markets in Oligopoly. The Warwick Economics Research Paper Series 260, University of Warwick,
Department of Economics.
Segal, I.R., and Whinston, M.D. (2000). Exclusive Contracts and Protection of Investments. The RAND
Journal of Economics, 31(4), 603-633.
Segal, I.R., and Whinston, M.D. (2000). Naked Exclusion: Comment. American Economic Review
90(1), 296-309.
Shapiro, C. (1999). Exclusivity in Network Industries. George Mason Law Review, 7(3), 673–83.
Unilateral Conduct Working Group (2008). Report on Predatory Pricing. International Competition
Network.
Vasconcelos, H. (2015). Is exclusionary pricing anticompetitive in two-sided markets? International
Journal of Industrial Organization, 40, 1–10.
Weyl, E. G. (2010). A price theory of multi-sided platforms. The American Economic Review, 100(4),
1642-1672.
Werden, G.J. (2006). Identifying Exclusionary Conduct Under Section 2: The ‘No Economic Sense’
Test. Antitrust Law Journal 73, 422–25.
Wright, J. (2004). One-sided logic in two-sided markets. Review of Network Economics, 3(1).
1. Introduction
Two-sided platforms refer to situations where (at least) two distinct user groups (i.e. two
demands) interact with each other through a common platform and the participation of at
least one of these groups impact the value of participation for the other group(s).
Following Evans (2003), "a platform constitutes the set of the institutional arrangements
necessary to realise a transaction between two users groups". 1
Typically, these two distinct customer groups cannot contract directly. This is because the
transaction costs for customers individually reaching enforceable agreements are too
high. As a result, a third party usually creates a place or space – a platform – where the
different groups of consumers/users can get together. In such situations, the need to
convince agents to participate on all sides of the platform creates a so-called chicken-and-
egg problem, in that members of each group are willing to participate in the market
insofar as they expect many members from the other side to participate.
For a market to be considered two-sided, it has to do more than just allow two or more
groups "to connect or engage with each other". As expressed by Rochet and Tirole, "if the
analysis just stopped there, pretty much any market would be two-sided, since buyers and
sellers need to be brought together for markets to exists and gains of trade to be
realized." 2 Yet two-sided markets are characterised not only by the existence of cross-side
network effects/indirect network effects, but also by the feature that the platform can use
its fee/pricing structure to influence the volume of transactions between users. Rochet and
Tirole therefore define a two-sided platform as one in which the volume of transactions
between users depends on the structure and not only on the overall level of the fees
charged by the platform.
Multi-sided platforms are very common and are present in many markets including: stock
exchanges, internet portals, payment card systems, newspapers, television broadcasters,
directories, smartphones, mobile and fixed telecommunication networks and estate
agents. These examples cover very diverse industries affecting many different aspects of
consumers' lives. For antitrust authorities it is therefore essential to have a thorough
understanding of these platforms to properly enforce antitrust scrutiny.
*
Tommaso Valletti (Chief Economist, DG COMP and Professor of Economics at Imperial College
Business School), Andrea Amelio (Economist, European Commission) and Liliane Karlinger
(Senior Economist at the Chief Economist Team of the European Commission).
Two-sided markets are an area of considerable recent economic research in the field of
Industrial Organization. The paper does not intend to provide an exhaustive review of the
two-sided market literature. The aim of the paper is two-fold. First and foremost, it
focuses on the literature dealing with exclusionary pricing and discusses whether the
presence of indirect network externalities makes platforms more or less prone to adopt
exclusionary conducts. Often, in the public debate, it is advocated that multi-sided
platforms deserve a special (typically, more relaxed) scrutiny by antitrust authorities. 3
The result of our preliminary research is not in line with this conclusion. Similar
exclusionary behaviours taking place in single-sided markets also carry over to multi-
sided markets. This suggests that the typical tools that one applies in the analysis of
single-sided markets need not to be abandoned: it is enough to adapt them. Second, the
paper discusses policy aspects that are particularly relevant in the current discussion
about platform competition and on which more research would be desirable.
The views and comments put forward in this paper are intended to add to the ongoing
debate on platforms and cannot be read as providing guidance on the European
Commission's past or future assessment of competition cases involving multi-sided
platforms. Our contribution has more modest goals and its main purpose is to contribute
with some embryonic research grounded on economic principles to the discussion about
the likelihood of exclusionary practices in multi-sided markets.
Both model families, "divide-and-conquer" and limit pricing, rely on the presence of scale
economies to achieve foreclosure, but divide-and-conquer strategies require the existence
of multiple buyers who can be played off against each other, while limit pricing models
give a first-mover advantage to the incumbent in making its price (or output) choices in a
way that leaves no room for entrants to establish their business alongside the incumbent
in the market.
This paper builds on these seminal works and adapts these models to incorporate a multi-
sided logic. We study how the presence of externalities, typical of multi-sided platforms,
changes the incentive of an incumbent firm to undertake exclusion.
aggressive price offers. However, the lack of an installed base proves to be a serious
obstacle when competing against the incumbent: If the entrant wins the new generation of
group 2 buyers, it will still have to compete against the incumbent for group 1 buyers,
because the incumbent's platform has positive value for group 1 buyers thanks to the
presence of the installed base. Thus, the profits that the entrant can recover on side 1 are
capped by the presence of a competitive incumbent.
The same is not true if instead the incumbent manages to attract group 2 buyers. Then, the
entrant's platform is completely worthless to group 1 buyers, so that the incumbent is
effectively a monopolist on this group and can extract monopoly rents from them. The
incumbent can therefore afford to be very aggressive in the fight for group 2 buyers,
because it can expect to recover higher profits on the other side of the market.
The paper shows that exclusion of the entrant can arise for a broad range of parameters,
namely when the cost advantage enjoyed by the entrant is relatively low compared to the
importance of the installed base. However, this does not necessarily imply that exclusion
is inefficient. A reader who is familiar with the literature on naked exclusion may
erroneously conclude that the fact that the entrant can serve buyers at a lower cost than
the incumbent automatically implies that total welfare is maximised when the entrant
serves the buyers, so that any equilibrium in which instead the incumbent prevails is
necessarily inefficient.
However, this conclusion does not necessarily carry over to the case of two-sided
markets. Here, there is an additional effect of entry on total welfare which needs to be
considered, namely the cost of splitting the two generations of group 2 buyers, old and
new, across two different platforms. 11 This is inherently inefficient, because it deprives
the young generation of group 1 buyers of the benefit of the network externality exerted
by the old generation of group 2 buyers, and vice versa.
More specifically, when both generations of group 2 buyers reside on the same platform,
the network benefits enjoyed by any buyer on the other side of the same platform amount
to 𝑧𝑧(𝛽𝛽 𝐼𝐼 + 𝑁𝑁); if all group 1 buyers get to enjoy these network effects (recall that the total
population of group 1 buyers is 𝛽𝛽 𝐼𝐼 + 𝑁𝑁), the total benefit will be 𝑧𝑧(𝛽𝛽 𝐼𝐼 + 𝑁𝑁)(𝛽𝛽 𝐼𝐼 + 𝑁𝑁). If
instead the two generations are fragmented across the two platforms, then the incumbent
platform generates network benefits of 𝑧𝑧(𝛽𝛽 𝐼𝐼 )2 , while the entrant's platform generates
𝑧𝑧𝑧𝑧 2 . This is clearly smaller than the total benefits when both cohorts are on the same
platform, i.e. 𝑧𝑧(𝛽𝛽 𝐼𝐼 + 𝑁𝑁)2 , because the latter also generates network benefits across
cohorts, not just within cohorts.
Under the assumptions of the model, only the incumbent network can generate the full
network effects of 𝑧𝑧(𝛽𝛽 𝐼𝐼 + 𝑁𝑁)2 , while entry necessarily leads to suboptimal network
benefits of 𝑧𝑧(𝛽𝛽 𝐼𝐼 )2 + 𝑧𝑧𝑧𝑧 2 . Overall, this model therefore exhibits an intricate set of
externalities: (i) the network benefits running from side 2 buyers of any platform to its
side 1 buyers, (ii) the network benefits running from the old cohort of the incumbent's
platform to its buyers in the young cohort, and (iii) the "contracting externalities" running
from the new cohort buyers on side 2 to those on side 1, because the side 2 buyers' choice
of platform also determines the available options for side 1 buyers. It is therefore not at
all obvious how these three layers of externalities will play out when the incumbent
engages in divide-and-conquer type of pricing.
The paper shows that when exclusion occurs in this model, it is always socially optimal:
Exclusion will occur when the entrant's cost advantage is not sufficient to outweigh the
benefits from having both generations of buyers concentrated on the same platform; and
this is precisely the condition under which entry is not desirable from a social welfare
point of view either. Moreover, there are equilibria where the entrant prevails but which
are nonetheless inefficient; in other words, this model may exhibit excessive entry.
Two lessons can therefore be learned from this model. The first lesson is that divide-and-
conquer strategies may be successfully used also in two-sided markets. As in a standard
one-sided market, some buyers may not fully internalise the impact their supplier choices
have on the options available to other buyers in the market, and an incumbent may take
advantage of this fact to lock in one part of the market by making very aggressive offers
to the other side of the market, thus preventing potential entrants from gaining a toehold
in the market.
The second lesson is that the impact of exclusion on social welfare might be different in a
two-sided market from a one-sided market. In this particular setup, the existence of an old
cohort of buyers, who are locked in with the incumbent, generates welfare losses if the
new cohort is served by the entrant instead of the incumbent, so that network externalities
are not maximised. Policies such as a ban on below-cost pricing, which are aimed at
preventing inefficient exclusion, may end up favouring inefficient entry instead.
a smaller installed reader base than the incumbent, 𝑛𝑛𝑅𝑅 < 𝑛𝑛𝐼𝐼 , so that its newspaper
currently provides lower utility to readers than the incumbent's, but has the potential to
provide higher utility if it manages to attract the new cohort of readers:
𝑣𝑣𝑅𝑅 (𝑛𝑛𝑅𝑅 ) < 𝑣𝑣𝐼𝐼 (𝑛𝑛𝐼𝐼 ) but 𝑣𝑣𝑅𝑅 (𝑛𝑛𝑅𝑅 + 1) > 𝑣𝑣𝐼𝐼 (𝑛𝑛𝐼𝐼 + 1). (𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 1)
Likewise, as regards advertisers' valuation for the newspapers, denoted 𝑎𝑎𝑖𝑖 (∙), the rival
newspaper, given its current small reader base, provides lower utility than the incumbent,
but is more efficient in providing advertisement benefits, so that advertisers would prefer
the rival newspaper if it managed to attract the new cohort of readers:
𝑎𝑎(𝑛𝑛𝑅𝑅 ) < 𝑎𝑎𝐼𝐼 (𝑛𝑛𝐼𝐼 ) but 𝑎𝑎𝑅𝑅 (𝑛𝑛𝑅𝑅 + 1) > 𝑎𝑎𝐼𝐼 (𝑛𝑛𝐼𝐼 + 1). (𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 2)
Also assume that the network externalities (both own-group and cross-group) increase
with a newspaper's reader base, but at a less-than-proportional rate. In order to simplify
the exposition, while still showing the main insights, we will focus on the special case
where the incumbent newspaper has fully exhausted all network effects, while the rival
newspaper still benefits from additional readers on both sides of its platform. In other
words, the readers' utility from reading the incumbent newspaper, vI, is unaffected by
whether or not the newspaper manages to attract the new cohort of readers of size 1, and
the same is true for advertisers:
𝑣𝑣𝐼𝐼 (𝑛𝑛𝐼𝐼 ) = 𝑣𝑣𝐼𝐼 (𝑛𝑛𝐼𝐼 + 1) and 𝑎𝑎𝐼𝐼 (𝑛𝑛𝐼𝐼 ) = 𝑎𝑎𝐼𝐼 (𝑛𝑛𝐼𝐼 + 1).
This assumption allows us to simplify our notation, by denoting respectively as v and a
the (constant) value to the readers and the advertisers when joining the incumbent
platform. Instead, variables with an overbar refer to the entrant when it manages to attract
the new cohort, while variables with an underbar refer to the opposite case when it fails to
do so. Hence we can restate our initial conditions as:
𝑣𝑣 > 𝑣𝑣 > 𝑣𝑣, (𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 1′ )
𝑎𝑎 > 𝑎𝑎 > 𝑎𝑎. (𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 2′ )
We will also make the simplifying assumption that both the cost of providing an ad, and
of providing the reader access to the newspaper, is zero.
Consider the following sequence of moves: first, the two newspapers compete for the new
cohort of readers by setting a uniform cover price for the newspaper, denoted 𝑝𝑝𝐼𝐼𝑟𝑟 , 𝑝𝑝𝑅𝑅𝑟𝑟 , and
then, they compete for the new cohort of advertisers by setting a uniform price per ad,
denoted 𝑝𝑝𝐼𝐼𝑎𝑎 , 𝑝𝑝𝑅𝑅𝑎𝑎 . 14
We can therefore apply backward induction to analyse which newspaper will prevail.
Clearly, at the second stage, competition for advertisers will depend on the outcome of
the first stage, i.e. whether it was the incumbent or the rival who managed to attract the
new readers. We consider each case in turn.
(2a) If the new cohort of readers bought I's newspaper at stage 1, then I's reader base is of
size 𝑛𝑛𝐼𝐼 + 1, which provides benefits of size a to advertisers, while R's reader base
remains at level 𝑛𝑛𝑅𝑅 , yielding lower benefits of 𝑎𝑎 to advertisers. The advertisers will
compare the net utility they are offered by I, namely 𝑎𝑎 − 𝑝𝑝𝐼𝐼𝑎𝑎 , to the net utility offered by
R, i.e. 𝑎𝑎 − 𝑝𝑝𝑅𝑅𝑎𝑎 , and will place their ads in I's newspaper whenever:
𝑎𝑎 − 𝑝𝑝𝐼𝐼𝑎𝑎 ≥ 𝑎𝑎 − 𝑝𝑝𝑅𝑅𝑎𝑎 .
Given that this is the last stage of the game, the lowest price R will be willing to offer its
advertisers is zero, so that I wins the advertisers with a positive price of
𝑝𝑝𝐼𝐼𝑎𝑎 = 𝑎𝑎 − 𝑎𝑎,
which leaves advertisers with a net utility of 𝑎𝑎 − 𝑝𝑝𝐼𝐼𝑎𝑎 = 𝑎𝑎.
(2b) If instead the new cohort of readers bought R's newspaper at stage 1, so that R has a
large reader base of size 𝑛𝑛𝑅𝑅 + 1 and provides a high utility of 𝑎𝑎 to advertisers, the latter
will prefer I's newspaper whenever:
𝑎𝑎 − 𝑝𝑝𝐼𝐼𝑎𝑎 ≥ 𝑎𝑎 − 𝑝𝑝𝑅𝑅𝑎𝑎 .
In this case, Bertrand competition among I and R will drive I's price offer down to zero,
and R wins the advertisers with a positive price of
𝑝𝑝𝑅𝑅𝑎𝑎 = 𝑎𝑎 − 𝑎𝑎,
which leaves advertisers with a net utility of 𝑎𝑎 − 𝑝𝑝𝑅𝑅𝑎𝑎 = 𝑎𝑎.
Let us now turn to competition for readers in stage 1. Recall that we assumed that readers
are indifferent as to how many advertisers any of the newspapers will attract at stage 2,
i.e. they only care about the newspaper's reader base, and its cover price. Thus, if they opt
for I's newspaper, the latter will have a reader base of size 𝑛𝑛𝐼𝐼 + 1, which provides net
benefits of 𝑣𝑣 − 𝑝𝑝𝐼𝐼𝑟𝑟 to readers; if instead they decide to buy R's newspaper, the latter will
have a reader base is of size 𝑛𝑛𝑅𝑅 + 1, which provides net benefits of 𝑣𝑣 − 𝑝𝑝𝑅𝑅𝑟𝑟 to readers.
Readers thus buy from I whenever:
𝑣𝑣 − 𝑝𝑝𝐼𝐼𝑟𝑟 ≥ 𝑣𝑣 − 𝑝𝑝𝑅𝑅𝑟𝑟 .
To see which of the two newspapers can make the more competitive offer to win the
readers, first note that their aggregate profits over the two periods, when successful in
period 1 (and ignoring discounting across the two periods), are:
Π𝐼𝐼 = 𝑝𝑝𝐼𝐼𝑟𝑟 + 𝑝𝑝𝐼𝐼𝑎𝑎 = 𝑝𝑝𝐼𝐼𝑟𝑟 + 𝑎𝑎 − 𝑎𝑎
Π𝑅𝑅 = 𝑝𝑝𝑅𝑅𝑟𝑟 + 𝑝𝑝𝑅𝑅𝑎𝑎 = 𝑝𝑝𝑅𝑅𝑟𝑟 + 𝑎𝑎 − 𝑎𝑎.
(1a) Consider first the scenario where I wins period 1 competition for readers. Bertrand
style competition between I and R ensures that the lowest price R is willing to offer is the
one that would drive its aggregate profits down to zero:
Π𝑅𝑅 = 0 → 𝑝𝑝𝑅𝑅𝑟𝑟 = −(𝑎𝑎 − 𝑎𝑎).
If I wants to match R's offer to win the readers in period 1, it has to offer:
𝑣𝑣 − 𝑝𝑝𝐼𝐼𝑟𝑟 = 𝑣𝑣 − 𝑝𝑝𝑅𝑅𝑟𝑟 → 𝑝𝑝𝐼𝐼𝑟𝑟 = 𝑣𝑣 − 𝑣𝑣 − (𝑎𝑎 − 𝑎𝑎).
Note that, given our assumptions on the parameters, this price is necessarily negative
(which is equivalent to being below marginal cost in this model, as the latter was assumed
to be zero). In other words, the incumbent can only attract readers by subsidising their
consumption.
At this price, I can break even whenever:
Π𝐼𝐼 = 𝑝𝑝𝐼𝐼𝑟𝑟 + 𝑝𝑝𝐼𝐼𝑎𝑎 = 𝑣𝑣 − 𝑣𝑣 − (𝑎𝑎 − 𝑎𝑎) + 𝑎𝑎 − 𝑎𝑎 > 0.
(1b) If the above break-even condition is not satisfied, i.e. if instead 𝑣𝑣 + 𝑎𝑎 − 𝑎𝑎 < 𝑣𝑣 +
𝑎𝑎 − 𝑎𝑎, then I will prefer to lose readers to R, so that R will make the sales to them at the
lowest price I is willing to offer, namely:
Π𝐼𝐼 = 0 → 𝑝𝑝𝐼𝐼𝑟𝑟 = −�𝑎𝑎 − 𝑎𝑎�.
Thus, a ban on negative prices would be an efficient policy tool to prevent exclusionary
pricing in this model. As argued above, exclusion is socially inefficient in this model,
because the entrant is more efficient than the incumbent at providing utility to both
readers and advertisers, provided it can attract both sides of the new cohort of consumers;
we also showed that whenever the incumbent instead manages to attract the readers
(which implies that all advertisers will then turn to the incumbent as well), this requires
the incumbent to set negative prices to the readers. It therefore follows that a ban on
negative prices will ensure that all instances of inefficient exclusion are ruled out.
Note, however, that this policy would not make everyone better off: whenever exclusion
would have occurred absent this ban on negative (i.e. below-cost) prices, buyers on the
reader side of the market will now pay a higher cover price, or, more precisely, they will
lose the subsidy they would have received from the incumbent. Advertisers instead will
benefit from this policy, because they obtain a larger net benefit in case the entrant
prevails.
In this standard setting, the analysis of Dixit (1980) is applied, with the important
difference that competition is in prices, not in quantities. One platform is considered to be
the incumbent and there is another platform that, if it decides to enter, will have to bear an
entry cost of K. This entry cost impacts negatively the expected profit of the new entrant.
The existence and the size of K is public information and therefore the incumbent
platform can take advantage of it. The incumbent has thus the option to either
accommodate entry becoming a Stackelberg leader, or to exclude entry and enjoy
monopolistic profits, albeit under the constraint that its output must be high enough (i.e.
its price must be low enough) to not leave any room for an entrant to cover its fixed cost
of entry. 16 Dixit (1980) shows that above a certain level of K, the incumbent has the
incentive to exclude the new entrant by expanding its capacity to a point where
production of the entry-deterring output level becomes a credible threat. In the setting of
the paper where platforms compete by setting prices, the analysis shows that the
incumbent has the same incentive to exclude the new entrant by decreasing prices.17
The introduction of the indirect network externalities does not change the basic intuition
identified in Dixit (1980), so that even platforms find it profitable to exclude entry. In the
following, the basic results of the analysis are derived and presented. 18
By assuming symmetric indirect network externalities, i.e. a1 = a2 = a > 0, and solving
the basic strategic game as described above, it is possible to derive the equation below
that identifies the difference between the profit of the incumbent from exclusion and from
accommodating entry.
1 − 𝛼𝛼
∆𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 = �32�𝐾𝐾/(1 − 𝛼𝛼) − 25�.
8
By solving the equation for ∆𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 = 0, one can find the critical threshold level for the entry
cost, K*(a), above which the incumbent prefers to exclude rather than accommodate the
entrant. This threshold depends on the intensity of the externality, α.
625(1 − 𝛼𝛼) 25
𝐾𝐾 ∗ (𝛼𝛼) = < (1 − 𝛼𝛼).
1024 16
By studying the function, it is straightforward to see that the higher the externality, the
lower K*(a). This implies that for any given K, the strategy of exclusion (i.e. lowering
the prices) becomes more attractive for the incumbent when indirect externalities are
stronger. Figure 1 below is another way of presenting the result, where the shaded area
represents the parameter region in the α-K space where the incumbent has the incentive to
exclude entry. The dark blue middle line represents the threshold level K*(a). Below this
line, the fixed cost of entry, K, is too low to make it worthwhile for the incumbent to deter
entry; the incumbent would rather accommodate the entrant and enjoy duopoly profits,
because deterrence through low prices would be too costly.
The yellow upper line represents the second threshold for K, namely the level at which
entry is "blockaded": when K exceeds the entrant's duopoly profits in the accommodating
scenario, it is never profitable for a competitor to enter because its expected profit will
never be positive. 19 This value also decreases with the level of the externality, meaning
that as the externalities become more intense, a monopoly is more and more likely to
arise even without any need for the incumbent to put an exclusionary strategy in place.
The blue wedge in Figure 1 is the most interesting from a policy point of view, because it
is the parameter region where the entrant would enter absent the strategic foreclosure by
the incumbent, but the incumbent finds it profitable to foreclose. It is possible to observe
that for a given K the presence of indirect network externalities makes the strategy of
foreclosure more attractive for the entrant. At the same time, we see from Figure 1 that
for any given level K, it is also more likely that entry will be blockaded, i.e. that entrants
will not find it profitable to enter even though the incumbent sets its prices in good faith,
i.e. in a way that is compatible with accommodation.
By relaxing the hypothesis of symmetry between the parameters capturing the indirect
network externalities, it is still possible to derive the equation below that identifies the
difference between the profit of the incumbent from exclusion and from accommodating
entry (and thus the strategic incentive of the incumbent to foreclose entry). Relaxing this
assumption is quite crucial given that arguably, the different sides of platforms typically
show different degrees of externality, which will not change simultaneously.
2(𝛼𝛼1 + 𝛼𝛼2 )2 + 𝛼𝛼1 𝛼𝛼2 − 9
∆𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 = −(2 − 𝛼𝛼1 − 𝛼𝛼2 ) + + (𝛼𝛼1 + 𝛼𝛼2 )�2(2 − 𝛼𝛼1 − 𝛼𝛼2 )𝐾𝐾
4(2 + 𝛼𝛼1 + 𝛼𝛼2 )
In this framework, solving for ∆𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 = 0 and finding the analytical expression of
K*(a1,a2) is more complex. The assessment is therefore done numerically, fixing one
parameter capturing the externality, here a1, while letting K and a2 vary. 20 Figure 2
shows the results of this exercise. The graph confirms the existence of the critical
threshold K*(a1,a2), above which exclusionary strategies become attractive for the
incumbent. 21 Moreover, it is also possible to observe that if there is a positive shock to
either of the parameters a1 or a2, the area where exclusionary strategies are desirable for
the incumbent expands. In other words, even for lower entry cost K it is still profitable for
the incumbent to price low in order to prevent the entrance of a competitor.
This allows concluding that it is enough to have a strong externality on one side of the
platform to make exclusion more attractive for the incumbent. A preliminary assessment
of these results suggests that the incumbent has the possibility to exclude entry on either
of the two sides. This can be consistent with the fact that the two sides are
interchangeable, and so the incumbent will always charge the lower price on the side of
the market where it is least costly to do so. Therefore, this might explain why the
structure of the network externalities across the two sides of the market does not seem to
matter, but only their overall intensity.
Figure 2. The critical threshold level K*(a1,a2) when network externalities are asymmetric
across the two sides of the platform
The application of these results should not be limited to the case of entry deterrence as
described above. It is also conceivable to interpret the entry cost parameter K as a
financial shock that can reduce the profitability of the follower. By giving this
interpretation to K, the results of the model take the flavour of financial predation.
Observing the financial shocks of the rival, the incumbent is taking advantage of these
financial fragilities of the rival and decides to decrease its price in order to make it
unprofitable for the rival to remain active in the market. This behaviour of the predator is
incentive compatible given the new structure of costs of the prey. The analysis above
seems to suggest that, in the presence of indirect network externalities, platforms are even
more prone to pursue predatory strategies of the kind described above.
All in all, the extensions of two strands of the literature on exclusionary practices are
consistent with indirect network externalities making it more likely for the incumbent to
engage in exclusionary behaviour. Moreover, it is enough to have an increase in the
indirect network externalities on at least one side of the platforms to make exclusionary
strategies more attractive to the incumbent.
3. Policy
The increasing importance of platforms in the current economy has raised several policy
debates. In this section, we select those that are the most relevant to the European
Commission and on which more research should be focused in order to come to a solid
understanding.
that predatory strategies typically observed in the context of standard markets carry over
to markets exhibiting indirect network externalities.
It is, however, important to stress that an analysis of indirect network externalities should
be part of the antitrust assessment. The typical tools applied in the analysis of single-sided
markets need not be abandoned but it is crucial to adapt them in order to capture the
specificity of platforms. One example of this effort is described in Behringer and
Filistrucchi (2015). 23 In order to evaluate predatory strategies of platforms, they propose
an augmented Areeda-Turner test that encompasses the presence of indirect network
externalities. By applying this test to two real-life examples they obtain two interesting
results. The first result shows that false positives might occur by applying a one-side test
in a context of indirect network externalities. This is indeed a call for using the right tool
when assessing platform competition. Their second result shows that a false negative
might also occur by applying a one-side test. This last result is thus also consistent with
the presence of predatory strategies performed by multi-sided platforms. This empirical
evidence in turn supports the position of maintaining an unchanged scrutiny of antitrust
authority for multi-sided platforms.
In conclusion, there seems to be convincing evidence that suggests that price structures
due to indirect network externalities can be used in a predatory fashion. Above-cost
predation is also possible if predation means sacrificing short-run profits to weaken rivals
and doing so in a way that lowers welfare. In this framework, predation can be hard to
detect: a standard price-cost test will not be reliable because there are non-predatory
reasons to price below cost; 24 and using the exit of rivals as indicator is not a sufficiently
solid standard of proof either, because the market may also tip absent predation. We can
thus conclude that the standard tools of antitrust analysis need to be adapted to the context
of two-sided markets to avoid false positives and false negatives alike.
Business asymmetries
A topic that seems to attract significant attention is asymmetric competition between the
advertising supported business model (i.e. multi-sided platform) and the subscriber-based
business model (traditional company). The asymmetry in the business models has a direct
repercussion on the competition for customers. One possible scenario faced by consumers
is that free products offered by the platform will compete with the products offered at a
positive price by the traditional company. In these circumstances it is often the case that
products show some degree of differentiation, either horizontal or vertical. Therefore it is
likely that the obvious effect of customers consuming the free product can be
significantly mitigated. However, from a policy perspective it might be important to
understand whether events like market exits by traditional businesses, which are most
likely to be displaced, should trigger antitrust intervention.
This is an open question and so far little effort has been put into trying to formally
understand the welfare implications of the competition between these two business
models.
by antitrust authority. However, the risk of a platform leveraging market power in one
market into adjacent markets should not be underestimated. Recent research, like Choi
and Jeon (2016), 26 focuses on the use of anticompetitive tying in order to overcome price
constraints, i.e. impossibility to charge negative prices. Price rigidities can be the result of
several factors, including the fear of triggering antitrust investigations for predatory
pricing. What the paper then suggests is that anticompetitive tying and predation are
interchangeable strategies. More attention therefore has to be put on tying and more
generally leveraging given that it can mask anticompetitive entry.
4. Conclusions
The aim of the paper is two-fold. First, it has a research objective as it extends two strands of the
literature about exclusionary pricing to the framework of indirect network externalities and platform
competition. Our preliminary results show that traditional exclusionary practices carry over to platform
competition and in some circumstances indirect network externalities accentuate the incentive to
foreclose by incumbents. Second, it also discusses some of the main policy topics that are currently
discussed in the public domain, complemented with some topics that so far have received little attention
despite their relevance.
Notes
1 Evans, David. “The Antitrust Economics of Multi-Sided Platform Markets.” Yale Journal on
Regulation 20 (2003): 325-82.
2 Jean-Charles Rochet and Jean Tirole. "Two-sided markets: a progress report." RAND Journal of
Economics 37, 3, (2006): 646.
3 See, for instance, Evans, David and Richard Schmalensee. “Industrial Organization of Markets
with Two-Sided Platforms.” Competition Policy International 3, 1 (2007): 27.
4 Segal, Ilya R., and Michael D. Whinston. "Naked exclusion: comment." The American Economic
Review 90, 1 (2000): 296-309.
5 Dixit, Avinash. "The Role of Investment in Entry-Deterrence." The Economic Journal 90, 357
(1980): 95-106.
6 Rasmusen, Eric B., J. Mark Ramseyer, and John S. Wiley Jr. "Naked exclusion." The American Economic
Review (1991): 1137-1145.
7 Segal, Ilya R., and Michael D. Whinston. "Naked exclusion: comment." The American Economic
Review 90, 1 (2000): 296-309.
8 Vasconcelos, Helder. "Is exclusionary pricing anticompetitive in two-sided markets?"
International Journal of Industrial Organization 40 (2015): 1-10.
9 The results of this paper are qualitatively similar if instead network externalities run in both
directions.
10 Subsidising consumption on one side of the platform makes sense whenever a platform wishes to
build an installed base of users on one side, so as to make its platform more attractive for users on
the other side, who will then be willing to pay a positive price to join. In this sense, platforms may
apply pricing practices that resemble divide-and-conquer strategies even without any exclusionary
intention, in particular whenever the network externalities are two-sided. In the present model,
however, a monopolist would not need to subsidise consumption on any side of the market: any
price slightly below r will ensure that group 2 buyers want to participate in this market, so that the
platform will also attract group 1 buyers and can charge positive prices to them.
11 Recall the assumption that the old buyers are locked in with the incumbent and cannot switch
platform in case of successful entry.
12 Fumagalli, Chiara, and Massimo Motta. "A simple Theory of Predation." The Journal of Law and
Economics 56, 3 (2013): 595-631.
13 In fact, their paper discusses the applicability of the general results to two-sided markets in a
subsection.
14 Competing first for readers, rather than advertisers, is a natural modelling choice here, given that
advertisers care about the readers anyone of the two newspapers manages to attract, while readers
are indifferent about the platform picked by the advertisers.
15 Armstrong, Mark. "Competition in two‐sided markets." The RAND Journal of Economics 37, 3
(2006): 668-691.
16 It is worth mentioning that the incumbent, under the monopoly scenario, serves the entire set of
customers, which in the Hotelling model amounts to a demand equal to 1, i.e. 100% of the
population.
17 One way for the incumbent to achieve commitment to such a limit price is to sign long-term
contracts with consumers which explicitly exclude price increases over the time horizon relevant
for entry.
18 Note that the demand functions in this model also depend on the differentiation parameters in each
Hotelling market, t1 and t2. For the purpose of this exercise and in order to simplify mathematical
expressions, the analysis sets t1 = t2 = 1. This means that the degree of differentiation of the two
markets involved are symmetric. For the purpose of the analysis this is irrelevant. On a more
technical ground, this assumption implies that the necessary and sufficient condition to have a
market sharing equilibrium becomes 4 - ( a1 + a2)2 > 0 which also implies that 2 - a1 - a2 > 0.
25
19 The expression of the threshold value for "blockaded" entry is 𝐾𝐾 = (1 − 𝛼𝛼).
16
Part V. Efficiencies
Abstract
Stock exchanges are platforms operating in multi-sided markets. Mergers between stock
exchanges can produce significant efficiency benefits, some of which can accrue directly
or indirectly to the users of the integrated exchange: intermediaries (brokers and
dealers), final investors and issuers (listed companies). In particular, stock exchange
integration can reduce the implicit costs of trading by increasing market liquidity. In this
paper we investigate the liquidity implications of the integration of Euronext’s cash
market. We find that the series of cash mergers that led to the creation of Euronext had a
positive impact on liquidity – namely, on bid-ask spreads, volatility and traded volume.
This exercise illustrates how past mergers can be used to assess empirically the potential
efficiencies resulting from mergers between platforms operating in multi-sided markets.
1. Introduction
1
Enrique Andreu is a Senior Vice President at Compass Lexecon. Jorge Padilla is Senior
Managing Director at Compass Lexecon, Research Fellow at CEMFI (Madrid) and teaches
competition economics at the Barcelona Graduate School of Economics (BGSE) and the Toulouse
School of Economics (TSE). This paper was prepared for presentation at the OECD Competition
Group Workshop Rethinking the use of traditional antitrust enforcement tools in multi-sided
markets to be held in Paris in June 2017. We wish to thank participants at the OECD workshop for
their comments and suggestions. This paper updates and upgrades using publicly available data the
econometric work we performed in the context of the Deutsche Börse / NYSE Euronext merger.
The opinions in this paper are our own and do not necessarily represent the views of Compass
Lexecon’s clients or other Compass Lexecon economists.
sided) markets need not apply in multi-sided markets. 3 In particular, mergers between
platforms competing in multi-sided markets need not be anticompetitive. First, the merger
may be welfare enhancing even when it leads to higher post-merger prices for both sides
of the market because users on either side of the merged platform will benefit from
increased access to a greater pool of users on the other side of platform. 4 Second, the
merger may even result in a reduction in prices since the merged platform may internalise
the cross-group externalities between the merging platforms: if the merging platforms
become “interoperable”,5 then each of the merging platforms will lower prices to benefit
from the increase in demand on the other platform. 6
The empirical evidence of the price and welfare effects of mergers in two-sided markets
is sparse. Some of these studies have focused on media markets, since those are
considered to be good examples of two-sided markets. Chandra and Collard-Wexler
(2009) investigated the price effects using data on a series of large merger in the
Canadian newspaper industry in the late 1990s. 7 Chandra and Collard-Wexler employed
difference-in-difference and difference-in-difference matching methods to compare price
changes in newspapers which change hands with those that did not and found that these
mergers did not lead to higher prices either for subscribers or advertisers. More recently,
Jeziorski (2014) examines the effects of mergers in the U.S. radio industry. 8 He finds that
they increase listener welfare marginally but have a more significant negative welfare
effect on advertisers.
A few authors have conducted post-mortem econometric analysis of mergers (i.e. ex-post
merger evaluations) among stock exchanges in order to assess their potential efficiencies,
if any. Stock exchanges are widely considered to be multi-sided markets. 9 Stock
exchange integration may in principle increase welfare by increasing market liquidity
and, hence, reducing the implicit costs of trading. The reduction of implicit costs may in
particular result from a reduction of bid-ask spreads or lower price volatility (because a
larger and more stable order flow reduces the noise induced by individual orders). There
are numerous mechanisms through which stock exchange mergers can increase liquidity
and decrease users’ implicit costs. 10 A merger between exchanges will increase liquidity
if it helps intermediaries to defray the costs of access to the trading platform and of
maintaining a continuous market presence. Standardised access to market data, indices
and post-trading services helps also the liquidity of integrated cash markets. Also
harmonised trading functionality, rules and regulations will reduce the regulatory costs of
trading in different markets. In addition, liquidity will increase if the merger reduces
adverse-selection costs, due to the presence of informed traders. This will happen if the
merger has a positive impact on trading activity and the additional order flow comes
mainly from uninformed traders or elicits more aggressive competition between informed
ones. A stock exchange merger may also increase liquidity (and lead to lower bid-ask
spreads) if it reduces the inventory-holding costs of market makers. This is because the
merger is likely to make the order flow more predictable and lower the costs of
rebalancing market-makers’ inventories after the execution of large orders. Finally,
liquidity may increase (and bid-ask spreads may fall) because the merger is likely to
induce entry by market professionals operating elsewhere, as a result e.g. of harmonised
rules and admission criteria, and thereby lead to greater competitive pressure both in
quote-setting and in brokerage fees.
In the U.S., Arnold et al. (1999) studied the effects on liquidity of three successive
mergers between regional U.S. stock exchanges in the 1940’s and 1950’s. 11 They found
that the bid-ask spreads of merged exchanges were narrower than those on the remaining
exchanges. The trend towards an efficient, consolidated capital markets infrastructure is
more recent in Europe than in the United States, but shows similar benefits. Pagano and
Padilla (2005b) investigated the liquidity effects resulting from the integration of the
French, Belgian, Dutch and Portuguese stock exchanges between September 2000 and
November 2003. 12 These mergers led to the creation of Euronext. This sequence of
mergers provides an extremely valuable natural experiment for the purposes of estimating
the liquidity effects of cash exchange mergers. First, the multi-stage nature of the
Euronext integration process – with three sequential mergers – makes it possible to better
identify the liquidity impact of stock exchange mergers, as it allows the empirical
estimation to deal more rigorously with spurious correlation. Second, since the timing of
the three mergers was predetermined at the outset and there were no departures from the
merger plan, there should be no concerns about reverse causality.
Pagano and Padilla (2005b) found that the creation of Euronext led to a reduction in the
bid-ask spreads of the large-cap securities traded in Paris, Brussels, Amsterdam and
Lisbon. They also found that the integration of those exchanges also led to an increase in
traded volume and a reduction in volatility for those stocks. Nielsson (2009) also
examined the liquidity effects of the Euronext integration process. 13 Unlike Pagano and
Padilla (2005b), he analysed the impact of the merger on the liquidity of all firms’ stocks
listed in the Paris, Brussels, Amsterdam and Lisbon exchanges and not just large caps.
Nielsson found that the Euronext mergers increased the liquidity and, therefore, reduced
the implicit trading costs of large caps. However, he found no statistically significant
effect of the merger on small and medium caps.
A difficulty with both studies is that their datasets are relatively limited and, in particular,
the duration of the post-merger period is short. Pagano and Padilla (2005b) only had data
until December 2004 – one year after the last integration event; Nielsson (2009) only
until 2006. This raises the concern that the effects that they attribute to the Euronext
mergers may not have been properly identified. Since the mergers took place at the time
of the collapse of the dot.com bubble (2000-2002) and the recession of the European
economy (2000-2001) and the U.S. economy (2002-2004), the post-merger increase in
liquidity documented in these papers may simply reflect the growth of trading volumes in
the aftermath of these crises.
Distinguishing between the liquidity effect of the mergers and these crises requires data
on a longer post-merger period than that used in Pagano and Padilla (2005b) and Nielsson
(2009). A longer post-merger period could help identify the effect of the mergers
correctly because, unlike the effect of the crises, the liquidity impact of a stock exchange
merger should be long lasting: the reduction in access costs, adverse selection costs,
inventory costs, and the increase in the strength of competition among intermediaries
resulting from the merger will likely persist indefinitely.
This paper thus revisits the analysis conducted by Pagano and Padilla (2005b) using data
from December 2000 to December 2010 to test whether, other things equal, the mergers
that led to the creation of Euronext had a long-lasting effect on bid-ask spreads, volatility
and volume. Expanding the post-merger period requires controlling for the important
changes in European cash trading that took place after the creation of Euronext. Most
importantly, the Markets in Financial Instruments Directive (MiFID) led to the entry of
new trading platforms (MTFs), which in a short time captured a significant market share
and is likely to have had an effect on liquidity. In addition, Euronext implemented a tick
size change in 2007, which is also likely to have had an impact on market liquidity.
Our results confirm Pagano and Padilla’s conclusions and show that the impact on market
liquidity that they identified is long lasting, as one would expect if those effects were
indeed caused by the creation of Euronext. Like them, we find that the creation of
Euronext increased the liquidity of the merging exchanges. This led to a reduction in the
bid-ask spreads and historical volatility of large-cap securities traded in Paris, Brussels,
Amsterdam and Lisbon. The creation of Euronext also resulted in an increase in traded
volume. These results are not only economically meaningful and statistically significant,
they are also robust and unlikely to be explained by omitted variables and reverse
causality (endogeneity).
This paper also investigates the potential liquidity impact of the merger between Euronext
and the NYSE Group (NYSE). This merger took place in April 2007. Because, unlike the
Euronext mergers, it did not involve the integration of the trading and clearing platforms
of the merging parties, we would expect the merger to have no impact on liquidity. This
is indeed what the data shows. We believe these results are consistent with our findings
on the creation of Euronext and serve to confirm them, since testing for the liquidity
impact of the merger between Euronext and the NYSE Group (NYSE) amounts to
performing a “placebo test.”
The remainder of this paper is structured as follows. Section 2 describes the integration
process that led to the creation of Euronext and the subsequent changes to the industry. In
Section 3 we investigate the impact of Euronext integration process on bid-ask spreads
using different data sources and econometric models. Section 4 presents several
robustness tests: using alternative integration dates, different measures of liquidity
(volatility and traded volume) and alternative controls. In Section 5, we analyse the
impact of the merger between NYSE and Euronext. Section 6 discusses the causal
interpretation of the results in Sections 3 to 5. Finally, Section 7 concludes with some
more general comments about the assessment of efficiencies in horizontal mergers in
multi-sided industries. All tables and figures described in the text can be found in the
annexes to the paper.
The creation of Euronext in September 2000 resulted in the integration of the French,
Belgian, Dutch, and Portuguese stock exchanges into a single trading and clearing
platform. Prior to the creation of Euronext, there were four separate trading and three
separate clearing platforms (Portugal had no CCP). Since November 2003, the users of
the Paris, Brussels, Amsterdam, and Lisbon exchanges have operated on a single trading
platform and a single clearing platform.
The integration of the cash markets that formed Euronext proceeded in stages. First, the
trading platform of the Paris market –the NSC system– became the platform for the other
three cash markets. In May 2001, the Brussels exchange migrated its trading platform to
the NSC system. Amsterdam followed suit in October 2001. Cash trading fees were
harmonised across Amsterdam, Brussels and Paris. The Lisbon exchange migrated to the
NSC system in November 2003.
The exchanges that form Euronext also integrated their clearing platforms. That process
took place in parallel, but with some delay, relative to the integration of the cash trading
platforms. The Paris market adopted the externally sourced Clearing 21 system in
September 2000. The Brussels cash market was migrated to the Clearing 21 system in
March 2002, while Amsterdam migrated in October 2002. Clearing operations across the
three locations were consolidated into Clearnet SA. Clearing in Lisbon was created in
November 2003.
Following Pagano and Padilla (2005b) our analysis considers the liquidity impact of the
migration of cash trading and clearing on the Amsterdam, Brussels, Paris and Lisbon cash
markets onto common trading and clearing platforms. However, we have also analysed
the potential effect of the integration of the trading platforms as a robustness test.
Since the creation of Euronext in November 2003 several events are likely to have
impacted the liquidity of the securities traded in Euronext. We highlight (and control for)
two such events in this paper. First, in November 2007 competitive trading platforms
known as multilateral trading facilities (or MTFs) entered the European cash trading
market. They have grown rapidly since then. MTFs are trading systems that make cash
instruments from different exchanges or sources available for trading. Currently there are
over five different pan-European blue-chip MTFs operating in Europe, the largest of
which is Chi-X. Second, in 2007 the Euronext tick size (the minimum price increment at
which trades may be made) was reduced and this is likely to have had an impact on
liquidity. Previous empirical literature has found that a reduction in the tick size leads to a
bid-ask spread reduction. This is because investors are able to tighten their quotes when
the minimum price increment becomes smaller.
In this section we analyse the impact of the creation of Euronext on bid-ask spreads using
standard multiple regression techniques. Bid-ask spreads are the most-often used
indicator of cash trading liquidity. We first describe the bid-ask spread data we use, the
methodology employed and our main results.
Econometric methodology
In order to assess the impact of the creation of Euronext on market liquidity, we conduct a
multi-stage before and after analysis around the three key integration milestones: (1) the
integration of the clearing and trading functions of the Paris and Brussels stock exchanges
in March 2002; (2) the integration of the Amsterdam trading and clearing system into
Euronext in October 2002; and (3) the integration of the Lisbon stock exchange in
November 2003. That is, we compare the daily bid-ask spreads defined above after each
of the integration dates with the same daily bid-ask spreads before integration. Our basic
models relate our bid-ask spread measure with an integration dummy that equals 1 after
integration and 0 before integration.
To isolate the impact of the creation of Euronext on bid-ask spreads, we control for other
factors that may explain differences in bid-ask spreads over time and across securities
traded on the different exchanges. This requires using multiple regression techniques. 16
We include in our basic regression model security fixed effects, a measure of market
volatility and indicator variables for relevant macroeconomic, political, and regulatory
events. 17 These variables take into account that differences in bid-ask spreads between
securities may persist over time and that bid-ask spread fluctuations may be driven by
factors other than the integration. We also included controls for volume and volatility
from non-Euronext exchanges, domestic GDP per capita, the volume traded at MTFs, and
an indicator variable to capture the effect of Euronext’s tick change in 2007.
More formally, we estimated a security-level panel-data model using Bloomberg’s daily,
security-level bid-ask spread data and a security-level panel-data model using Euronext’s
security-level weighted average bid-ask spread data. We used a panel data approach
because this allows us to (1) avoid complex aggregation issues, (2) estimate the impact of
integration on all the exchanges of the Euronext platform in a single regression, and (3)
obtain estimates for the impact of integration in the various exchanges of Euronext that
can be readily compared.
Our basic security-level panel-data model can be formally written as,
yit = a + β1 Integrationit + β 2 Z it + β 3 X it + ηi + λt + e it
where: 18
• yit is the natural logarithm of the bid-ask spread of security i at period t.
• Integrationit is a dummy variable that takes the value of 1 for any security i and
period t after the integration of the trading and clearing platforms of the exchange
where security i is traded, and 0 otherwise. The sign of the coefficient of the
integration dummy characterises the relation between the bid-ask spread and the
creation of Euronext. A negative sign would indicate that bid-ask spreads declined
(and thus liquidity increased) as a result of the creation of Euronext.
• Zit is a vector of variables that control for other determinants of the liquidity of the
market. We included: the (20-day) historical volatility of the FTSE100 and DAX
indices (source: Bloomberg); the traded volume on the Frankfurt exchange
(source: Deutsche Börse); a tick size dummy, which takes a value of 1 after 2007
and is equal to 0 before then; 19 the per capita GDP of each of the countries with
Euronext exchanges (source: Eurostat); and the volume traded at MTFs (source:
Bloomberg). 20
The first two controls are meant to capture common trends of a global or pan-European
nature that could have affected the bid-ask spreads of Euronext’s large-cap securities and
that have nothing to do with the process of formation of Euronext. We would expect to
find a positive relation between bid-ask spreads and the volatility of the FTSE100 and
DAX indices, and a negative relationship between bid-ask spreads and the volume traded
on the Frankfurt exchange, a proxy for market growth. The tick size dummy is meant to
capture the impact on liquidity of the reduction in tick size implemented in all Euronext
exchanges in 2007. We introduce a GDP per capita measure because it may drive the
volume traded in each of the four Euronext exchanges. Increases in volume may have an
impact on liquidity and, hence, on bid-ask spreads. Finally, the volume traded at MTFs
may also have had an impact on the liquidity of the regulated exchanges that integrated
Euronext and thus on their bid-ask spreads. The sign of this variable may be negative (if it
proxies an increase in overall traded volume) or positive (if MTFs divert significant
liquidity out of the regulated exchanges).
• Xit is a vector of dummy variables that controls for some relevant economic and
political events. These events may have affected the liquidity of the Euronext
exchanges before and after the creation of Euronext (see the list of events
considered in Annex 2).
• η i is a vector of fixed effects: one per security. These dummies are introduced to
capture security-specific factors that may influence the liquidity of those
securities and that are independent of the process of integration.
• λt is a vector of monthly fixed effects. These dummies control for monthly-
specific shocks that may have affected liquidity in the stock exchange markets
under consideration and that have nothing to do with the process of integration.
• ε it denotes the standard statistical error.
This panel data model is estimated using ordinary least squares (OLS). We calculate
robust standard errors, clustering at the security level to allow for heteroskedasticity and
autocorrelation of the errors. Because the creation of Euronext was triggered by an
exogenous policy decision, we can safely place a causal interpretation on the econometric
estimates for the Integration dummy, provided other relevant changes in the economic
environment are controlled for. Causality runs from the integration events to the
estimated changes in liquidity. 21
Econometric results
This Section reports the results of the econometric estimation of the basic model
described above. 22 Table D.1 (Annex D) presents the results of our empirical analysis
using the Bloomberg measure of the daily bid-ask spreads of the securities included in the
main indices of the Amsterdam, Brussels, Lisbon and Paris stock exchanges. Column (1)
describes the impact of integration on the average bid-ask spread of the securities listed in
those exchanges. Columns (2) and (3) replicate the model in Column (1) controlling for
changes in the volatility of the DAX index 23 and in the volume traded on the Frankfurt
exchange. These controls are added to take account of potential liquidity trends that are
unrelated to the creation of Euronext. Column (4) also includes the effect of the Euronext
tick size reduction in 2007, and Columns (5) and (6) control in addition for the entry and
growth of MTFs and changes in per capita GDP in each of the four Euronext countries.
Once again, the logic behind these controls is to isolate the liquidity impact of the
creation of Euronext and avoid confounding it with the effects of these other variables.
The regressions in all columns include monthly dummies, security dummies and
dummies controlling for a few salient events that may have affected the behavior of bid-
ask spreads in the relevant exchanges. Across all specifications we find that the average
bid-ask spreads of the securities included in the main indices of the Paris, Brussels,
Amsterdam and Lisbon stock exchanges fell as a result of the creation of Euronext. This
effect is statistically significant in Columns (1) to (3) which include controls for changes
in the volatility of the securities listed in the DAX and traded volumes on the Frankfurt
exchange. Both of these control variables have the expected sign (positive for volatility
and negative for traded volumes) and are statistically significant.
However, when we include the indicator for the change in the tick size, the growth of
MTFs and GDP per capita as controls, the estimated impact of the integration continues
to be negative but is no longer statistically significant. Each of the additional control
variables has an impact on bid-ask spreads that is consistent with finance and economic
theory. The tick size reduction led to a reduction in the bid-ask spread as did the
increasing volume of trades on MTFs and the GDP per capita.
The loss of statistical significance of the integration dummy in Columns (4) to (6) should
not be a matter of concern, as is most likely due to multicollinearity. 24 A way to test for
this is to disaggregate the impact of integration to consider the changes in bid-ask spreads
resulting from each of the three steps in the creation of Euronext. The sequential nature of
Euronext’s integration process allows the empirical analysis to control in part for spurious
correlations that could bias the estimation of the impact of integration.
This can be done in two alternative ways. First, we define three different integration
dummies: Phase 1 (which equals 1 for any security i and period t after the integration of
Brussels with Paris if security i is traded either in Brussels or Paris, and 0 otherwise);
Phase 2 (which equals 1 for any security i and period t after the integration of Amsterdam
if security i is traded either in Brussels, Paris or Amsterdam, and 0 otherwise), and Phase
3 (which equals 1 for any security i and period t after the integration of Lisbon if security
i is traded either in Brussels, Paris, Amsterdam or Lisbon, and 0 otherwise). This
approach makes it easier to disentangle the liquidity impact of the integration from the
effects of other unrelated changes that may have occurred post-merger.
Table D.2 (Annex D) presents the results of this alternative modelling approach. We find
a statistically significant and material decline of the average bid-ask spreads of the
securities included in the main indices of the Paris, Brussels, Amsterdam and Lisbon
stock exchanges as a result of the creation of Euronext under all specifications (i.e., in
Columns (1) to (6)). In particular, Phase 1 and Phase 3 are statistically significant even
when we include the indicator for the change in the tick size, the growth of MTFs and
GDP per capita as controls. Note also that the estimated coefficients of all of the control
variables continue to be statistically significant and have signs which are consistent with
economic and finance theory. The estimated coefficients of the three phase variables in
Columns (5) and (6) imply that the creation of Euronext has led to a non-transitory
reduction in bid-ask spreads of approximately 50%. Columns (5) and (6) also show that
the largest decline occurred when the Brussels and Paris exchanges were integrated. The
subsequent integration of the Amsterdam and Lisbon exchanges had a smaller impact on
the bid-ask spreads.
An alternative approach is to replace the integration dummy in Table D.1 by four
country-specific integration dummies: Paris (which takes a value of 1 for any security i
traded in Paris and period t after the integration of Brussels with Paris, and 0 otherwise);
Brussels (which takes a value of 1 for any security i traded in Brussels and period t after
the integration of Brussels with Paris, and 0 otherwise); Amsterdam (which takes the
value of 1 for any security i traded in Amsterdam and period t after the integration of
Amsterdam in the Euronext platform, and 0 otherwise); and Lisbon (which takes the
value of 1 for any security i traded in Lisbon and period t after the integration of Lisbon
in the Euronext platform, and 0 otherwise). These variables may capture the differential
effect of integration on the Paris, Brussels, Amsterdam and Lisbon markets. This is
important since the liquidity impact of Euronext’s creation may have been different in the
various exchanges, reflecting differences, among other things, in size, depth and breadth.
Table D.3 (Annex D) presents the results of this alternative modelling approach. We find
that the negative effect of integration on the average bid-ask spread identified in Table
D.2 is largely driven by the effect of integration on the liquidity of the securities listed in
the CAC 40 and the BEL 20. Note that this is true even after controlling for the change in
the tick size in 2007, the entry and subsequent growth of MTFs, and GDP per capita.
From Columns (4) and (5), we observe that the average bid-ask spread of the securities in
the CAC 40 fell approximately 59% as a result of integration. The bid-ask spread of the
securities in the BEL20 fell approximately 25%. These effects are material and
statistically significant. The results for Amsterdam are not so clear cut, however. We find
that while the impact of integration on the bid-ask spreads of the securities listed in the
Amsterdam index was to reduce spreads, the relation is not statistically significant.
Finally, we note that integration led to an increase of the bid-ask spreads in Lisbon. 25
4. Robustness tests
In this section we report the results of several robustness tests. 26 First, we repeat the
various analyses in Section III using different integration milestones: we employ as
relevant cut offs the dates at which the trading platforms of the different exchanges were
integrated, as opposed to the dates at which their clearing platforms were integrated.
Second, we investigate the impact of the creation of Euronext on alternatives measures of
liquidity: volatility and volume. Third, we present the results of our econometric analysis
adding the price of the securities and market capitalisation as additional explanatory
variables. The results of these robustness checks confirm the main findings of Section III,
namely that the creation of Euronext increased the liquidity of the merging exchanges
and, therefore, reduced the implicit costs of trading. The increase in liquidity is reflected
in lower bid-ask spreads even after using alternative integration dates. It also results in
lower volatility and higher volume.
statistically significant. While the coefficient for the Phase 2 dummy is not statistically
significant, this does not necessarily imply that the merger of Amsterdam had no liquidity
impact. This is because the integration of Amsterdam into Euronext took place only five
months after the integration of Paris and Brussels and, as a result, our model may not be
able to estimate with sufficient precision the effect of Phase 2 of the integration process.
The largest decline in bid-ask spreads occurred at the time of the integration of the Paris
and Brussels exchanges.
Impact on volatility
Other things equal, markets with a large number of traders (i.e., thick markets) are less
volatile than thinner markets. Larger and more stable order flows in thick markets reduce
the noise induced by individual orders, since they tend to average out and, therefore, to
exert less pressure on prices. Moreover, thick markets have a tighter bid-ask spread and
thus the “bid-ask price bounce” induced by large orders is smaller. In addition, the price
concession necessary to execute a large order is smaller in thick markets because there is
a greater likelihood of finding a trading counterparty. For all these reasons, a merger
between exchanges that results in a thicker market should lead also to lower volatility. 29
Following Pagano and Padilla (2005b), we analyse the impact of integration on volatility
using 20-day historical volatility data for the securities included in the CAC 40, BEL 20,
AEX, and PSI. The source of this data is Bloomberg. We have data on a daily basis for
the period between 3 January 2000 and 31 December 2010. The number of observations
is 365,335. We use the same econometric methodology employed in section III. We
include all controls used above plus, in addition, the historical volatility of the CAC 40,
BEL 20, AEX, and PSI indices. We introduce this variable as an additional way to isolate,
to the extent possible, the effect of integration on volatility from other confounding
factors such as shocks to the global or European economy, or shocks that are
idiosyncratic to the exchanges considered but are not related to the integration process.30
Table E.2 presents the results of the econometric analysis of the impact on volatility of
the creation of Euronext using Integration as the dependent variable. We find that the
historical volatility of the large-cap securities traded in Paris, Brussels, Amsterdam and
Lisbon fell as a result of the creation of Euronext. This effect is statistically significant in
all specifications. It is also material from an economic viewpoint. According to the
estimation in Columns (6) and (7), which are the ones that we prefer given that they
include all relevant controls, volatility fell on average approximately 9% once the
integration of the four exchanges was completed.
Impact on volume
We also analyse the impact of the creation of Euronext on traded volume, measured for
the purposes of this analysis by the number of shares traded for the securities included in
the CAC 40, BEL 20, AEX, and PSI. The number of shares traded for these securities has
been obtained from Bloomberg. We have data on a daily basis for the period between 3
January 2000 and 31 December 2010. The number of observations is 382,146. Once
again we use the same methodology and the same control variables as in section 3.
Table E.3 presents the results of the econometric analysis. We find that traded volume in
the Paris, Brussels, Amsterdam and Lisbon exchanges increased as a result of the creation
of Euronext. This effect is statistically significant. It is also material from an economic
viewpoint. According to the estimations in Columns (5) and (6), which are the ones that
we prefer given that they include all relevant controls, volume increased by
approximately 25% in the period after the integration. A thicker market (i.e., a market
with more traders and greater traded volume) is a more liquid market. 31 Therefore, our
findings on the effect of the Euronext mergers on traded volume confirm our previous
results on bid-ask spreads and volatility: the Euronext mergers increased liquidity.
The NYSE- Euronext merger took place in April 2007. In this section, we explore the
impact of the NYSE-Euronext merger in April 2007 on the bid-ask spreads of the
Euronext exchanges. We would expect to find no statistically significant impact of the
merger on bid-ask spreads, given that the merger did not involve the integration of the
trading and/or clearing platforms of NYSE and Euronext. We use the same methodology
and the same control variables as in section 3, except that we add a dummy variable
(NYSE merger) that takes the value of one after 4 April 2007 and zero otherwise. We
restrict the sample to three years before and after the NYSE-Euronext merger to isolate
the impact of this event. Estimation results are presented in Table F.1 (Annex F). 33 We
find no effect of the merger once we control for the tick size reduction and the entrance of
MTFs. This implies that the liquidity effects that we have identified in this paper are not
due to a mere change in ownership but rather the consequence of the integration of the
underlying trading and clearing platforms of the merging exchanges.
Our econometric results show that exchange mergers that result in the integration of the
underlying trading and clearing platforms produce material liquidity effects, which reduce
the implicit cost of trading and thus ought to benefit market participants and offset the
potentially adverse impact of the merger on exchange fees (i.e. the explicit cost of
trading). However, before these results can be extrapolated to other mergers across stock
changes, a few comments on endogeneity and causality are in order.
First, the multi-stage nature of the Euronext integration process –with three sequential
mergers– makes it possible in our opinion to identify the liquidity impact of stock
exchange mergers. In particular, we agree with Pagano and Padilla (2005b) and Nielsson
(2009) that the staggered introduction of merger events across the four participating
exchanges allows the empirical estimation to deal more rigorously with spurious
correlation.
Second, to the best of our knowledge, the timing of the four mergers was predetermined
at the outset and there were no departures from the merger plan. Therefore, we believe
that there are no reasons to doubt the causal interpretation we have given to our
regression results. We find no reason to be concerned about reverse causality and
spurious correlation. There is no evidence that liquidity increased in the years prior to the
merger. There is also no evidence that a third omitted factor triggered the mergers and the
change in liquidity.
In particular, there is no evidence of a downward trend in the data, 34 as we proceed to
discuss. Note first that if there was an omitted trend, we would expect the residuals of a
model like those reported in Section 3 which excluded the integration dummies,
Integrationit, to show a continuously declining trend. However, as shown in Figure G.1 in
Annex G, this is not the case. (Dotted lines show the integration dates.) The time
evolution of the residuals (the bid-ask component not explained by all relevant controls
except the merger integration indicators) reveals that the main trends present in the
original data are properly captured by the control variables included in the models
presented in Section III. The difference between actual and predicted bid-ask spreads
follow a stable pattern during the pre-merger period and then during the post-merger.
However, it falls significantly below their pre-merger mean value once exchanges are
integrated into Euronext, indicating the presence of a level shift following the mergers.
Thus, the analysis of the difference between actual and predicted bid-ask spreads does not
show a downward trend prior to the integration of the Euronext exchanges. On the
contrary, the graphical analysis show a different evolution of the bid ask spreads before
(fluctuating around a stable mean) and after (decreasing over time) the integration of the
exchanges. In other words, it appears to require the introduction of step functions like the
integration dummies in the models of Section III.
We have tested for the existence of an omitted trend indicating a general trend toward
increasing liquidity that was unrelated to the Euronext integration more formally. In
particular, we have modified the models in Section III above by removing the integration
dummies, Integrationit, and introducing instead a series of quarter-year fixed effects for
each exchange. That is, for each exchange, we have introduced as many dummy variables
as quarters in the sample; each quarter dummy takes a value of 1 in that quarter and 0
otherwise. We have then estimated the exchange by exchange, instead of pooling all
exchanges into a single regression. The results are consistent with those in Section III
above. 35 Figure G.2 in Annex G shows the results of the estimated quarterly fixed effects
of the modified. None of these estimates shows a downward trending pattern, which
further confirms that the effects of the Euronext mergers are not picking up an omitted
long-term trend.
Third, the mergers that gave rise to Euronext involved exchanges with a similar structure.
They were all hybrid markets with limit order book emphasis. They had all introduced an
order driven, electronic, continuous market at the time of the merger. Therefore, the
liquidity impact we have identified cannot be attributed to changes in market structure.
Fourth, while Amsterdam had demutualised in 1997, a few years before its integration in
Euronext, the other three mergers demutualised as part of the merger. This may raise
doubts about whether the liquidity impact that we have attributed to the mergers is instead
the effect of demutualisation. We do not believe that this alternative explanation is
correct. While the economic literature has found that demutualisation is likely to have an
impact on the liquidity of the newly demutualised exchange, 36 there is no evidence of
effects across exchanges. However, we find evidence that each sequential merger had a
positive impact on the liquidity of the Paris and Brussels exchanges. We also find
evidence that the integration process increased liquidity in Amsterdam, which as noted
before had demutualised prior to the merger.
Fifth, our results on the liquidity effect of the merger cannot be attributed to tick size
harmonisation. This is for the following reasons: (a) the Paris exchange – which benefited
most from the merger in terms of increased liquidity – introduced new tick sizes in 1999,
before the start of the integration; (b) Amsterdam and Lisbon aligned their tick sizes to
the Paris model in dates that did not overlap with the merger dates; (c) while changes in
tick size may have an impact on the liquidity of the stock exchange where the tick change
occurs, we find evidence that each merger had an impact on the other participating
exchanges; and (d) we find a permanent effect of each phase of integration even after
controlling for the tick change of 2007, which does not square with an alternative
interpretation that attributes the liquidity effects of the creation of integration to the
process of tick size harmonisation that run parallel to the integration process.
Sixth, the Euronext integrations pre-dated the introduction of MTFs. The effects that the
model identifies are therefore clearly attributable to the mergers. However, our
econometric analysis of efficiencies in the cash market also indicates that the introduction
and subsequent growth of MTFs led to a statistically significant increase in liquidity, i.e. a
reduction in bid-ask spreads and volatility and an increase in traded volume. There is no
contradiction between the positive effects of exchange fragmentation, which results in
increasing competition for a given security and, hence, a possible reduction in the explicit
cost of trading and integration, and exchange integration, where different securities traded
in different venues are pooled together and distributed to a wider set of traders.
7. Concluding remarks
Stock exchanges are platforms that co-ordinate traders willing to sell with those willing
to buy. Because traders in either side can be inefficiently rationed and not all traders
provide good matchings to their potential counterparts, stock exchanges are clear-cut
examples of multi-sided platforms. Therefore, mergers between exchanges need not lead
to higher prices and, on the contrary, are likely to benefits all, or some, of the market
participants by increasing liquidity. In this paper we have shown that the efficiency
effects of mergers in multi-sided markets, such as stock exchanges, are not merely
theoretical and can be assessed using standard econometric techniques. Note finally that
while some of these efficiencies might have been clawed back in the form of higher
trading fees, a significant share of them must have been appropriated by the users of
Euronext. This would be true in a one-sided market, since not even a monopolist would
normally be able to fully appropriate these demand-side efficiencies, but is even more so
in multi-sided markets, where post-merger price competition may be fiercer.
This paper illustrates how the efficiencies created by the merger of two or more platforms
can be estimated empirically ex post. The so-called post-mortem approach to the
quantification of horizontal merger efficiencies requires estimating the link between past
concentrations and platform measures which directly or indirectly capture the magnitude
of the access and usage externalities benefiting users on both sides of the platform. As we
have seen above, this is a complex exercise because it requires controlling for possible
confounding factors, taking into account the potential endogeneity of the mergers,
performing robustness tests, etc. Extrapolating the results of ex-post analyses of this sort
when reviewing a new merger is also challenging, since not all mergers are equal and the
market context where the past mergers took place may not resemble that applying to the
new transaction. However, none of this means that this approach has no value. The
alternative is to simulate the impact of the platform merger on users’ utilities or profits.
This requires estimating demand functions on both sides of the market, including direct
and indirect network effects, which is a much more complex exercise. Besides the usual
complexity in demand estimation, these simulations must take account of the degree of
interoperability across the merging platforms that existed pre-merger and of the
prevalence of multi-homing in one or more market sides .
Table A.1 provides descriptive statistics on the (normalised) bid-ask spreads provided by
Bloomberg.
Table A.1. Descriptive statistics for the Bloomberg-based bid-ask spreads of large caps in
Paris, Brussels, Amsterdam and Lisbon, 3 January 2000 – 31 December 2010here
Number of
Index observations Average Standard deviation Minimum Maximum
Amsterdam (AEX) 98,613 0.012 0.052 0 1.887
Belgium (BEL) 69,892 0.005 0.013 0 1.78
Paris (CAC) 114,825 0.002 0.003 0 0.479
Lisbon (PSI) 78,773 0.012 0.060 0 1.995
Source: Bloomberg e.
Figure A.1. Estimated year-month fixed effects – Bid-ask spread for AEX, BEL, CAC, and
PSI securities. Jan 2000-Dec 2010
-5
-5.5
Month-year fixed effects
Month-year fixed effects
-5.5
-6
-6
-6.5
-6.5
-7 -7
03jan2000
27apr2001
17apr2003
06apr2005
30aug2000
20aug2002
09aug2004
30jul2006
19jul2008
09jul2010
23dec2001
13dec2003
02dec2005
22nov2007
11nov2009
27mar2007
16mar2009
06mar2011
03jan2000
27apr2001
17apr2003
06apr2005
30aug2000
20aug2002
09aug2004
30jul2006
19jul2008
09jul2010
23dec2001
13dec2003
02dec2005
22nov2007
11nov2009
27mar2007
16mar2009
06mar2011
-5
Month-year fixed effects
Month-year fixed effects
-6.5
-5.2
-7
-5.4
-7.5
-5.6
-8
03jan2000
27apr2001
17apr2003
06apr2005
30aug2000
20aug2002
09aug2004
30jul2006
19jul2008
09jul2010
23dec2001
13dec2003
02dec2005
22nov2007
11nov2009
27mar2007
16mar2009
06mar2011
03jan2000
27apr2001
17apr2003
06apr2005
30aug2000
20aug2002
09aug2004
30jul2006
19jul2008
09jul2010
23dec2001
13dec2003
02dec2005
22nov2007
11nov2009
27mar2007
16mar2009
06mar2011
Note: The vertical lines indicate the different consolidation dates of the Paris, Brussels and Amsterdam
exchanges into Euronext. The first line shows the integration of the Paris and Brussels exchanges, the second
one the integration of the Amsterdam exchange and the third one, the integration of the Lisbon exchange..
Source: Bid-ask spreads calculated using price data provided by Bloomberg.
Table D.1. Impact of integration on the normalised bid-ask spreads of large caps in Paris,
Brussels, Amsterdam and Lisbon, 3 January 2000 – 31 December 2010
Ln (bid-ask spread) 1 2 3 4 5 6
Integration -0.590*** -0.530*** -0.119** -0.071 -0.068 -0.019
[0.000] [0.000] [0.024] [0.174] [0.193] [0.698]
DAX volatility (log of) 0.245*** 0.365*** 0.406*** 0.418*** 0.359***
[0.000] [0.000] [0.000] [0.000] [0.000]
Traded volume on the Frankfurt exchange -0.911*** -0.314*** -0.307*** -0.166***
(log of) [0.000] [0.000] [0.000] [0.000]
Tick change dummy -0.519*** -0.322*** -0.047
[0.000] [0.000] [0.590]
Per capita GDP -4.859***
[0.000]
MTF volume (log of) -0.012** -0.022***
[0.016] [0.000]
Constant -4.666*** -4.146*** 16.643*** 3.243*** 3.076*** 49.412***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Month fixed effects Yes Yes Yes Yes Yes Yes
Security fixed effects Yes Yes Yes Yes Yes Yes
Observations 362,103 357,936 332,831 332,832 332,831 332,831
R-squared 0.523 0.533 0.570 0.590 0.599 0.603
Notes:
(a) Robust p-value in brackets, clustered by security to allow for heteroskedasticity and autocorrelation within
securities. Estimation residuals are stationary according with the Fisher stationary test for panel regressions.
See Maddala, G.S. and S. Wu (1999) “A comparative study of unit root tests with panel data and a simplified
test”, Oxford Bulleting of Economics and Statistics, pp. 631-652.
(b) * significant at 10%; ** significant at 5%; *** significant at 1%.
(c)The sample is composed by 158 large caps. It includes securities that have composed the main index of the
Paris, Brussels, Amsterdam and Lisbon exchanges (the CAC 40, BEL20, AEX and the PSI) at any point
throughout the sample period.
(d) Bid-ask spreads calculated using price data provided by Bloomberg.
Table D.2. Impact of integration on the normalised bid-ask spreads of large caps in Paris,
Brussels, Amsterdam and Lisbon by phases, 3 January 2000 – 31 December 2010
Ln (bid-ask spread) 1 2 3 4 5 6
Phase 1 -0.344*** -0.356*** -0.14 -0.238** -0.241** -0.259**
[0.004] [0.003] [0.219] [0.031] [0.029] [0.016]
Phase 2 -0.155 -0.168 -0.187* -0.174* -0.175* -0.125
[0.160] [0.120] [0.050] [0.065] [0.062] [0.115]
Phase 3 -0.570*** -0.519*** -0.310*** -0.172*** -0.164*** -0.117**
[0.000] [0.000] [0.000] [0.000] [0.000] [0.016]
DAX volatility (log of) 0.109*** 0.225*** 0.306*** 0.317*** 0.307***
[0.000] [0.000] [0.000] [0.000] [0.000]
Traded volume on the Frankfurt
-0.642*** -0.177*** -0.173*** -0.110***
exchange
(log of) [0.000] [0.000] [0.000] [0.002]
Tick change dummy -0.461*** -0.332*** -0.180*
[0.000] [0.000] [0.053]
Per capita GDP -2.669**
[0.047]
MTF volume(log of) -0.008 -0.014**
[0.106] [0.018]
Constant -4.530*** -4.304*** 10.500*** 0.08 0.001 25.781*
[0.000] [0.000] [0.000] [0.921] [0.999] [0.054]
Month fixed effects Yes Yes Yes Yes Yes Yes
Security fixed effects Yes Yes Yes Yes Yes Yes
Observations 362,103 357,936 332,831 332,831 332,831 332,831
R-squared 0.558 0.563 0.597 0.606 0.606 0.607
Notes:
(a) Robust p-value in brackets, clustered by security to allow for heteroskedasticity and autocorrelation within
securities. Estimation residuals are stationary according with the Fisher stationary test for panel regressions.
See Maddala, G.S. and S. Wu (1999) “A comparative study of unit root tests with panel data and a simplified
test”, Oxford Bulleting of Economics and Statistics, pp. 631-652.
(b) * significant at 10%; ** significant at 5%; *** significant at 1%.
(c) The sample is composed by 158 large caps. In particular, we include securities that have composed the
main index of the Paris, Brussels, Amsterdam and Lisbon exchanges (the CAC 40, BEL20, AEX and the PSI)
at any point throughout the sample period.
(d) Bid-ask spreads calculated using price data provided by Bloomberg.
Table D.3. Impact of integration on the normalised bid-ask spreads of large caps in Paris,
Brussels, Amsterdam and Lisbon by exchange, 3 January 2000 – 31 December 2010
Ln (bid-ask spread) 1 2 3 4 5 6
Paris -1.119*** -1.073*** -0.586*** -0.594*** -0.592*** -0.561***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Brussels -0.757*** -0.724*** -0.247** -0.257** -0.256** -0.143
[0.000] [0.000] [0.017] [0.012] [0.012] [0.182]
Amsterdam -0.491*** -0.432*** -0.147 -0.085 -0.081 0.023
[0.003] [0.008] [0.306] [0.549] [0.563] [0.851]
Lisbon -0.127 -0.029 0.247*** 0.349*** 0.356*** 0.287***
[0.101] [0.722] [0.002] [0.000] [0.000] [0.000]
DAX volatility (log of) 0.272*** 0.380*** 0.426*** 0.440*** 0.389***
[0.000] [0.000] [0.000] [0.000] [0.000]
Traded volume on the Frankfurt exchange -0.900*** -0.262*** -0.253*** -0.146***
(log of) [0.000] [0.000] [0.000] [0.000]
Tick change dummy -0.553*** -0.324*** -0.11
[0.000] [0.000] [0.178]
Per capita GDP (log of) -3.800***
[0.001]
MTF volume (log of) -0.015*** -0.021***
[0.005] [0.000]
Constant -4.777*** -4.184*** 16.449*** 2.111*** 1.909** 38.133***
[0.000] [0.000] [0.000] [0.008] [0.016] [0.001]
Month fixed effects Yes Yes Yes Yes Yes Yes
Security fixed effects Yes Yes Yes Yes Yes Yes
Observations 362,103 357,936 332,831 332,831 332,831 332,831
R-squared 0.534 0.545 0.592 0.592 0.607 0.609
Notes:
(a) Robust p-value in brackets, clustered by security to allow for heteroskedasticity and autocorrelation within
securities. Estimation residuals are stationary according with the Fisher stationary test for panel regressions.
See Maddala, G.S. and S. Wu (1999) “A comparative study of unit root tests with panel data and a simplified
test”, Oxford Bulleting of Economics and Statistics, pp. 631-652.
(b) * significant at 10%; ** significant at 5%; *** significant at 1%.
(c) The sample is composed by 158 large caps. It includes securities that have composed the main index of
the Paris, Brussels, Amsterdam and Lisbon exchanges (the CAC 40, BEL20, AEX and the PSI) at any point
throughout the sample period.
(d) Bid-ask spreads calculated using price data provided by Bloomberg.
Table E.1. Impact of integration of trading platforms on the normalised bid-ask spreads of
large caps in Paris, Brussels, Amsterdam and Lisbon, 3 January 2000 – 31 December 2010
Ln (bid-ask spread) 1 2 3 4 5 6
Phase 1 -0.407*** -0.405*** -0.422*** -0.314*** -0.316*** -0.305***
[0.001] [0.001] [0.000] [0.004] [0.004] [0.006]
Phase 2 -0.059 -0.086 0.071 -0.098 -0.102 -0.089
[0.576] [0.400] [0.402] [0.252] [0.235] [0.270]
Phase 3 -0.657*** -0.610*** -0.414*** -0.264*** -0.256*** -0.179***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
DAX volatility (log of) 0.104*** 0.216*** 0.298*** 0.308*** 0.298***
[0.000] [0.000] [0.000] [0.000] [0.000]
DB volume (log of) -0.659*** -0.183*** -0.178*** -0.111**
[0.000] [0.000] [0.000] [0.020]
Tick change dummy -0.455*** -0.330*** -0.152
[0.000] [0.000] [0.101]
Per capita GDP -3.089**
[0.034]
MTF volume(log of) -0.008 -0.015**
[0.120] [0.015]
Month fixed effects Yes Yes Yes Yes Yes Yes
Security fixed effects Yes Yes Yes Yes Yes Yes
Constant -4.538*** -4.301*** 10.718*** 0.211 0.122 30.121**
[0.000] [0.000] [0.000] [0.745] [0.853] [0.032]
Observations 362,103 357,936 332,831 332,831 332,831 332,831
R-squared 0.557 0.561 0.596 0.605 0.605 0.606
Notes:
(a) Robust p-value in brackets, clustered by security to allow for heteroskedasticity and autocorrelation within
securities. Estimation residuals are stationary according with the Fisher stationary test for panel regressions.
See Maddala, G.S. and S. Wu (1999) “A comparative study of unit root tests with panel data and a simplified
test”, Oxford Bulleting of Economics and Statistics, pp. 631-652.
(b) * significant at 10%; ** significant at 5%; *** significant at 1%.
(c)The sample is composed by 158 large caps. It includes securities that have composed the main index of the
Paris, Brussels, Amsterdam and Lisbon exchanges (the CAC 40, BEL20, AEX and the PSI) at any point
throughout the sample period.
(d) Bid-ask spreads calculated using price data provided by Bloomberg.
Ln (volatility) 1 2 3 4 5 6 7
Integration -0.187*** -0.062** -0.122*** -0.071*** -0.085*** -0.088*** -0.085***
[0.000] [0.025] [0.000] [0.005] [0.001] [0.000] [0.001]
DAX volatility (log of) 0.596*** 0.589*** 0.576*** 0.561*** 0.557***
[0.000] [0.000] [0.000] [0.000] [0.000]
Index volatility (log of) 0.588***
[0.000]
Traded volume on the Frankfurt
exchange 0.211*** 0.012 -0.001 0.006
(log of) [0.000] [0.544] [0.979] [0.770]
Tick change dummy 0.177*** -0.114*** -0.100**
[0.000] [0.003] [0.029]
Per capita GDP (log of) -0.251
[0.632]
MTF volume(log of) 0.018*** 0.018***
[0.000] [0.000]
Constant -1.550*** -0.243*** -0.140*** -5.090*** -0.616 -0.338 2.075
[0.000] [0.000] [0.001] [0.000] [0.193] [0.481] [0.691]
Month fixed effects Yes Yes Yes Yes Yes Yes Yes
Security fixed effects Yes Yes Yes Yes Yes Yes Yes
Observations 365,335 363,248 365,335 336,252 336,252 336,252 336,252
R-squared 0.275 0.470 0.512 0.497 0.505 0.507 0.507
Notes:
(a) Robust p-value in brackets, clustered by security to allow for heteroskedasticity and autocorrelation within
securities. Estimation residuals are stationary according with the Fisher stationary test for panel regressions.
See Maddala, G.S. and S. Wu (1999) “A comparative study of unit root tests with panel data and a simplified
test”, Oxford Bulleting of Economics and Statistics, pp. 631-652.
(b) * significant at 10%; ** significant at 5%; *** significant at 1%
(c) The sample is composed by 158 large caps. In particular, we include securities that have composed the
main index of the Paris, Brussels, Amsterdam and Lisbon exchanges (the CAC 40, BEL20, AEX and the PSI)
at any point throughout the sample period.
(d) 20-Day historical volatility calculated using last prices provided by Bloomberg.
Table E.3. Impact of integration on number of shares traded for large caps in Paris,
Brussels, Amsterdam and Lisbon, 3 January 2000 – 31 December 2010
Ln (number of shares) 1 2 3 4 5 6
Integration 0.607*** 0.244*** 0.271*** 0.253*** 0.253*** 0.254***
[0.000] [0.001] [0.000] [0.000] [0.000] [0.000]
Per capita GDP (log of) 9.117*** 9.359*** 6.876*** 5.852*** 5.812***
[0.000] [0.000] [0.000] [0.005] [0.007]
Traded volume on the Frankfurt exchange (log of) 0.403*** 0.293*** 0.296***
[0.000] [0.000] [0.000]
DAX volatility (log of) 0.105*** 0.053* 0.030 0.032
[0.001] [0.068] [0.319] [0.270]
Tick change dummy 0.149 0.182
[0.119] [0.216]
MTF volume (log of) -0.002
[0.820]
Constant 9.373*** -83.758*** -86.048*** -69.931*** -57.031*** -56.677***
[0.000] [0.000] [0.000] [0.000] [0.006] [0.007]
Month fixed effects Yes Yes Yes Yes Yes Yes
Security fixed effects Yes Yes Yes Yes Yes Yes
Observations 382,146 382,146 377,403 349,152 349,152 349,152
R-squared 0.760 0.771 0.774 0.783 0.783 0.783
Notes:
(a) Robust p-value in brackets, clustered by security to allow for heteroskedasticity and autocorrelation within
securities. Estimation residuals are stationary according with the Fisher stationary test for panel regressions.
See Maddala, G.S. and S. Wu (1999) “A comparative study of unit root tests with panel data and a simplified
test”, Oxford Bulleting of Economics and Statistics, pp. 631-652.
(b) * significant at 10%; ** significant at 5%; *** significant at 1%.
(c) The sample is composed by 158 large caps. In particular, we include securities that have composed the
main index of the Paris, Brussels, Amsterdam and Lisbon exchanges (he CAC 40, BEL20, AEX and the PSI)
at any point throughout the sample period.
(d) Number of shares traded provided by Bloomberg.
The table below presents the results of the econometric analysis of the impact of the
integration by phases on the normalised bid-ask spreads when the daily market
capitalisation is included as an additional control.
Table E.4. Impact of integration on the normalised bid-ask spreads of large caps in Paris,
Brussels, Amsterdam and Lisbon, controlling for market capitalisation,
by phases, 3 Jan 2000 – 31 Dec 2010
Ln (bid-ask spread) 1 2 3 4 5 6
Phase 1 -0.227*** -0.215*** -0.094 -0.183*** -0.193*** -0.191***
[0.000] [0.001] [0.125] [0.002] [0.001] [0.001]
Phase 2 -0.402*** -0.400*** -0.390*** -0.372*** -0.384*** -0.389***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Phase 3 -0.327*** -0.353*** -0.198*** -0.078* -0.042 -0.045
[0.000] [0.000] [0.000] [0.054] [0.290] [0.308]
DAX volatility (log of) -0.055* 0.038 0.114*** 0.152*** 0.153***
[0.051] [0.123] [0.000] [0.000] [0.000]
Traded volume on the -0.439*** -0.028 -0.005 -0.011
Frankfurt exchange (log of) [0.000] [0.319] [0.847] [0.690]
Tick change dummy -0.410*** 0.143** 0.131**
[0.000] [0.026] [0.035]
Per capita GDP (log of) 0.234
[0.805]
MTF volume (log of) -0.036*** -0.035***
[0.000] [0.000]
Market capitalisation (log of) -0.618*** -0.626*** -0.604*** -0.592*** -0.612*** -0.614***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Constant 0.647* 0.602* 10.522*** 1.182** 0.868 -1.39
[0.064] [0.069] [0.000] [0.032] [0.119] [0.881]
Month fixed effects Yes Yes Yes Yes Yes Yes
Security fixed effects Yes Yes Yes Yes Yes Yes
Observations 347,837 345,605 329,145 329,145 329,145 329,145
R-squared 0.625 0.627 0.642 0.65 0.652 0.652
Notes: (a) Robust p-value in brackets, clustered by security to allow for heteroskedasticity and autocorrelation
within securities; (b) * significant at 10%; ** significant at 5%; *** significant at 1%; (c) the sample is
composed by 158 large caps. In particular, we include securities that have composed the main index of the
Paris, Brussels, Amsterdam and Lisbon exchanges (the CAC-40, BEL-20, AEX and the PSI) at any point
throughout the sample period; (d) all regressions include a set of single-day event dummies; (e) bid-ask
spreads calculated using price data provided by Bloomberg.
The table below presents the results of the econometric analysis of the impact of the
integration by exchanges on the normalised bid-ask spreads when the daily market
capitalisation is included as an additional control.
Table E.5. Impact of integration on the normalised bid ask spread of large caps in Paris,
Brussels, Amsterdam and Lisbon controlling for market capitalisation, by exchanges,
3 Jan 2000 – 31 Dec 2010
Ln (bid-ask spread) 1 2 3 4 5 6
Paris -1.046*** -1.041*** -0.700*** -0.702*** -0.702*** -0.697***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Brussels -0.639*** -0.636*** -0.294*** -0.301*** -0.299*** -0.285***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Amsterdam -0.630*** -0.616*** -0.367*** -0.303*** -0.302*** -0.288***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Lisbon 0.038 0.057 0.280*** 0.370*** 0.392*** 0.383***
[0.586] [0.427] [0.000] [0.000] [0.000] [0.000]
DAX volatility (log of) 0.062** 0.161*** 0.208*** 0.239*** 0.234***
[0.014] [0.000] [0.000] [0.000] [0.000]
Traded volume on the -0.636*** -0.072** -0.038 -0.026
Frankfurt exchange (log of) [0.000] [0.015] [0.206] [0.267]
Tick change dummy -0.494*** 0.147** 0.171***
[0.000] [0.021] [0.004]
Per capita GDP (log of) -0.48
[0.563]
MTF volume (log of) -0.041*** -0.041***
[0.000] [0.000]
Market capitalisation (log of) -0.675*** -0.663*** -0.613*** -0.594*** -0.617*** -0.614***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Constant 1.020*** 1.061*** 15.185*** 2.331*** 1.764*** 6.332
[0.000] [0.000] [0.000] [0.000] [0.003] [0.439]
Month fixed effects Yes Yes Yes Yes Yes Yes
Security fixed effects Yes Yes Yes Yes Yes Yes
Observations 347,837 345,605 329,145 329,145 329,145 329,145
R-squared 0.614 0.616 0.638 0.65 0.653 0.653
Notes: (a) Robust p-value in brackets, clustered by security to allow for heteroskedasticity and autocorrelation
within securities; (b) * significant at 10%; ** significant at 5%; *** significant at 1%; (c) the sample is
composed by 158 large caps. In particular, we include securities that have composed the main index of the
Paris, Brussels, Amsterdam and Lisbon exchanges (the CAC-40, BEL-20, AEX and the PSI) at any point
throughout the sample period; (d) all regressions include a set of single-day event dummies; (e) bid-ask
spreads calculated using price data provided by Bloomberg.
The table below presents the results of the econometric analysis of the impact of the
integration by phases on the normalised bid-ask spreads when the daily closing price of
each security is included as an additional control.
Table E.6. Impact of integration on the normalised bid-ask spreads of large caps in Paris,
Brussels, Amsterdam and Lisbon, controlling for daily closing price of the security,
by phases, 3 Jan 2000 – 31 Dec 2010
Ln (bid-ask spread) 1 2 3 4 5 6
Phase 1 -0.265*** -0.247*** -0.076 -0.176*** -0.186*** -0.186***
[0.000] [0.000] [0.229] [0.004] [0.002] [0.001]
Phase 2 -0.442*** -0.436*** -0.412*** -0.398*** -0.415*** -0.416***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Phase 3 -0.426*** -0.464*** -0.269*** -0.129*** -0.089** -0.090**
[0.000] [0.000] [0.000] [0.001] [0.011] [0.036]
DAX volatility (log of) -0.074*** 0.032 0.115*** 0.157*** 0.157***
[0.010] [0.226] [0.000] [0.000] [0.000]
Traded volume on the -0.534*** -0.061** -0.036 -0.038
Frankfurt exchange (log of) [0.000] [0.017] [0.165] [0.149]
Tick change dummy -0.468*** 0.165** 0.162**
[0.000] [0.012] [0.018]
Per capita GDP (log of) 0.056
[0.955]
MTF volume (log of) -0.041*** -0.041***
[0.000] [0.000]
Closing price of the security -0.558*** -0.572*** -0.574*** -0.573*** -0.601*** -0.601***
(log of) [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Constant -2.752*** -2.862*** 9.425*** -1.157** -1.614*** -2.16
[0.000] [0.000] [0.000] [0.039] [0.005] [0.828]
Month fixed effects Yes Yes Yes Yes Yes Yes
Security fixed effects Yes Yes Yes Yes Yes Yes
Observations 360,460 356,315 331,215 331,215 331,215 331,215
R-squared 0.607 0.611 0.635 0.645 0.647 0.647
Notes: (a) Robust p-value in brackets, clustered by security to allow for heteroskedasticity and autocorrelation
within securities; (b) * significant at 10%; ** significant at 5%; *** significant at 1%; (c) the sample is
composed by 158 large caps. In particular, we include securities that have composed the main index of the
Paris, Brussels, Amsterdam and Lisbon exchanges (the CAC-40, BEL-20, AEX and the PSI) at any point
throughout the sample period; (d) all regressions include a set of single-day event dummies; (e) bid-ask
spreads calculated using price data provided by Bloomberg.
This table presents the results of the econometric analysis of the impact of the integration
by exchanges on the normalised bid-ask spreads when the daily closing price of each
security is included as an additional control.
Table E.7. Impact of integration on the normalised bid ask spread of large caps in Paris,
Brussels, Amsterdam and Lisbon controlling for daily closing price of the security,
by exchanges, 3 Jan 2000 – 31 Dec 2010
Ln (bid-ask spread) 1 2 3 4 5 6
Paris -1.198*** -1.178*** -0.728*** -0.735*** -0.739*** -0.730***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Brussels -0.828*** -0.814*** -0.352*** -0.363*** -0.365*** -0.337***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Amsterdam -0.775*** -0.749*** -0.429*** -0.367*** -0.373*** -0.345***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Lisbon -0.068 -0.042 0.245*** 0.349*** 0.373*** 0.356***
[0.279] [0.523] [0.000] [0.000] [0.000] [0.000]
DAX volatility (log of) 0.079*** 0.181*** 0.228*** 0.262*** 0.251***
[0.002] [0.000] [0.000] [0.000] [0.000]
Traded volume on the -0.768*** -0.114*** -0.075*** -0.050**
Frankfurt exchange (log of) [0.000] [0.000] [0.007] [0.032]
Tick change dummy -0.566*** 0.166** 0.215***
[0.000] [0.011] [0.001]
Per capita GDP (log of) -0.925
[0.273]
MTF volume (log of) -0.047*** -0.048***
[0.000] [0.000]
Closing price of the security -0.595*** -0.582*** -0.568*** -0.569*** -0.602*** -0.597***
(log of) [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Constant -2.753*** -2.638*** 14.880*** 0.188 -0.575 8.261
[0.000] [0.000] [0.000] [0.745] [0.335] [0.327]
Month fixed effects Yes Yes Yes Yes Yes Yes
Security fixed effects Yes Yes Yes Yes Yes Yes
Observations 360,460 356,315 331,215 331,215 331,215 331,215
R-squared 0.590 0.593 0.628 0.644 0.647 0.648
Notes: (a) Robust p-value in brackets, clustered by security to allow for heteroskedasticity and autocorrelation
within securities; (b) * significant at 10%; ** significant at 5%; *** significant at 1%; (c) the sample is
composed by 158 large caps. In particular, we include securities that have composed the main index of the
Paris, Brussels, Amsterdam and Lisbon exchanges (the CAC-40, BEL-20, AEX and the PSI) at any point
throughout the sample period; (d) all regressions include a set of single-day event dummies; (e) bid-ask
spreads calculated using price data provided by Bloomberg.
The table below presents the results of the econometric analysis of the impact of the
integration by phases on the normalised bid-ask spreads when the market capitalisation
and the daily closing price of each security are included as additional controls.
Table E.8. Impact of integration on the normalised bid-ask spreads of large caps in Paris,
Brussels, Amsterdam and Lisbon, controlling for market capitalisation and closing price of
the security, by phases, 3 Jan 2000 – 31 Dec 2010
Ln (bid-ask spread) 1 2 3 4 5 6 7 8
Phase 1 -0.241** -0.259** -0.193*** -0.191*** -0.186*** -0.186*** -0.185*** -0.183***
[0.029] [0.016] [0.001] [0.001] [0.002] [0.001] [0.001] [0.001]
Phase 2 -0.175* -0.125 -0.384*** -0.389*** -0.415*** -0.416*** -0.403*** -0.410***
[0.062] [0.115] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Phase 3 -0.164*** -0.117** -0.042 -0.045 -0.089** -0.090** -0.051 -0.057
[0.000] [0.016] [0.290] [0.308] [0.011] [0.036] [0.175] [0.201]
DAX volatility 0.317*** 0.307*** 0.152*** 0.153*** 0.157*** 0.157*** 0.143*** 0.144***
(log of) [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Traded volume on the
Frankfurt -0.173*** -0.110*** -0.005 -0.011 -0.036 -0.038 -0.009 -0.017
Exchange (log of) [0.000] [0.002] [0.847] [0.690] [0.165] [0.149] [0.723] [0.522]
Tick change -0.332*** -0.180* 0.143** 0.131** 0.165** 0.162** 0.184*** 0.166***
Dummy [0.000] [0.053] [0.026] [0.035] [0.012] [0.018] [0.004] [0.006]
MTF volume -0.008 -0.014** -0.036*** -0.035*** -0.041*** -0.041*** -0.040*** -0.039***
(log of) [0.106] [0.018] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Per capita GDP -2.669** 0.234 0.056 0.337
(log of) [0.047] [0.805] [0.955] [0.709]
Market capitalisation -0.612*** -0.614*** -0.436*** -0.438***
(log of) [0.000] [0.000] [0.000] [0.000]
Last price of the -0.601*** -0.601*** -0.200* -0.200*
security (log of) [0.000] [0.000] [0.064] [0.064]
Constant 0.001 25.781* 0.868 -1.39 -1.614*** -2.16 0.05 -3.2
[0.999] [0.054] [0.119] [0.881] [0.005] [0.828] [0.947] [0.724]
Month fixed effects Yes Yes Yes Yes Yes Yes Yes Yes
Security fixed effects Yes Yes Yes Yes Yes Yes Yes Yes
Observations 332,831 332,831 329,145 329,145 331,215 331,215 327,529 327,529
R-squared 0.606 0.607 0.652 0.652 0.647 0.647 0.651 0.651
Notes: (a) Robust p-value in brackets, clustered by security to allow for heteroskedasticity and autocorrelation
within securities; (b) * significant at 10%; ** significant at 5%; *** significant at 1%; (c) the sample is
composed by 158 large caps. In particular, we include securities that have composed the main index of the
Paris, Brussels, Amsterdam and Lisbon exchanges (the CAC-40, BEL-20, AEX and the PSI) at any point
throughout the sample period; (d) all regressions include a set of single-day event dummies; (e) bid-ask
spreads calculated using price data provided by Bloomberg.
This table presents the results of the econometric analysis of the impact of the integration
by exchanges on the normalised bid-ask spreads when the market capitalisation and the
daily closing price of each security are included as additional controls.
Table E.9. Impact of integration on the normalised bid-ask spreads of large caps in Paris,
Brussels, Amsterdam and Lisbon, controlling for market capitalisation and closing price of
the security, by exchanges, 3 Jan 2000 – 31 Dec 2010
Ln (bid-ask spread) 1 2 3 4 5 6 7 8
Paris -0.592*** -0.561*** -0.702*** -0.697*** -0.739*** -0.730*** -0.713*** -0.708***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Brussels -0.256** -0.143 -0.299*** -0.285*** -0.365*** -0.337*** -0.316*** -0.302***
[0.012] [0.182] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Amsterdam -0.081 0.023 -0.302*** -0.288*** -0.373*** -0.345*** -0.331*** -0.318***
[0.563] [0.851] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Lisbon 0.356*** 0.287*** 0.392*** 0.383*** 0.373*** 0.356*** 0.385*** 0.376***
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
DAX volatility 0.440*** 0.389*** 0.239*** 0.234*** 0.262*** 0.251*** 0.233*** 0.228***
(log of) [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Traded volume on the
Frankfurt -0.253*** -0.146*** -0.038 -0.026 -0.075*** -0.050** -0.04 -0.029
exchange (log of) [0.000] [0.000] [0.206] [0.267] [0.007] [0.032] [0.159] [0.196]
Tick change dummy -0.324*** -0.11 0.147** 0.171*** 0.166** 0.215*** 0.184*** 0.207***
[0.000] [0.178] [0.021] [0.004] [0.011] [0.001] [0.004] [0.000]
MTF volume -0.015*** -0.021*** -0.041*** -0.041*** -0.047*** -0.048*** -0.045*** -0.045***
(log of) [0.005] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Per capita GDP -3.800*** -0.48 -0.925 -0.443
(log of) [0.001] [0.563] [0.273] [0.569]
Market capitalisation -0.617*** -0.614*** -0.467*** -0.464***
[0.000] [0.000] [0.000] [0.000]
Last price of the -0.602*** -0.597*** -0.170* -0.171*
security (log of) [0.000] [0.000] [0.084] [0.083]
Constant 1.909** 38.133*** 1.764*** 6.332 -0.575 8.261 1.053 5.271
[0.016] [0.001] [0.003] [0.439] [0.335] [0.327] [0.164] [0.500]
Month fixed effects Yes Yes Yes Yes Yes Yes Yes Yes
Security fixed effects Yes Yes Yes Yes Yes Yes Yes Yes
Observations 332,831 332,831 329,145 329,145 331,215 331,215 327,529 327,529
R-squared 0.607 0.609 0.653 0.653 0.647 0.648 0.651 0.651
Notes: (a) Robust p-value in brackets, clustered by security to allow for heteroskedasticity and autocorrelation
within securities; (b) * significant at 10%; ** significant at 5%; *** significant at 1%; (c) the sample is
composed by 158 large caps. In particular, we include securities that have composed the main index of the
Paris, Brussels, Amsterdam and Lisbon exchanges (the CAC-40, BEL-20, AEX and the PSI) at any point
throughout the sample period; (d) all regressions include a set of single-day event dummies; (e) bid-ask
spreads calculated using price data provided by Bloomberg.
Table F.1. Impact of the NYSE-Euronext merger on the normalised bid-ask spreads of large
caps in Paris, Brussels, Amsterdam and Lisbon, 2 January 2004 – 31 December 2010
Ln (bid-ask spread) 1 2 3 4 5 6 7
NYSE merger -0.351*** -0.471*** -0.328*** -0.039 -0.011 -0.004 0.012
[0.000] [0.000] [0.000] [0.327] [0.779] [0.924] [0.777]
DAX volatility 0.247*** 0.291*** 0.272*** 0.276*** 0.272*** 0.290***
(log of) [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
.5 .5
0 0
-.5 -.5
-1 -1
01dec2000
01dec2001
01dec2002
01dec2003
30nov2004
30nov2005
30nov2006
30nov2007
29nov2008
29nov2009
29nov2010
01dec2000
01dec2001
01dec2002
01dec2003
30nov2004
30nov2005
30nov2006
30nov2007
29nov2008
29nov2009
29nov2010
Paris, CAC Portugal, PSI
1.5
1.5
1
1
.5
.5
-.5
-.5
-1
01dec2000
01dec2001
01dec2002
01dec2003
30nov2004
30nov2005
30nov2006
30nov2007
29nov2008
29nov2009
29nov2010
01dec2000
01dec2001
01dec2002
01dec2003
30nov2004
30nov2005
30nov2006
30nov2007
29nov2008
29nov2009
29nov2010
Note: (i) the horizontal line represents the average residual before the integration of Paris; (ii) dotted vertical
lines shows dates of integration: Paris and Brussels on 1 March 2002; Amsterdam on 25 October 2002; and
Lisbon on 1 November 2003.
Table G.1. Regression results using a specification without integration variables and adding
quarter-year index fixed effects, Paris (CAC) securities only,
1 December 2000 – 31 December 2010
Table G.2. Regression results using a specification without integration variables and adding
quarter-year index fixed effects, Belgium (BEL) securities only,
1 December 2000 – 31 December 2010
Table G.3. Regression results using a specification without integration variables and adding
quarter-year index fixed effects, Amsterdam (AEX) securities only,
1 December 2000 – 31 December 2010
Table G.4. Regression results using a specification without integration variables and adding
quarter-year index fixed effects, Portugal (PSI) securities only,
1 December 2000 – 31 December 2010
.5
-.5
2000q4
2001q2
2001q4
2002q2
2002q4
2003q2
2003q4
2004q2
2004q4
2005q2
2005q4
2006q2
2006q4
2007q2
2007q4
2008q2
2008q4
2009q2
2009q4
2010q2
2010q4
AEX CAC BEL PSI
Note: (i) Estimated time fixed effects are fitted under specification 4 in the tables in Annex G; (ii) The sample
used begins on 1 December 2000 as volumes traded on the Frankfurt exchange are available from that date
on.
Notes
1 See, e.g. Rochet, J.C., and J. Tirole (2006) “Two-sided markets: a progress report,” RAND
Journal of Economics, 37(3): 645-667
2 Professors Hermalin and Katz note that in these markets the sum of the socially optimal prices for
the platform services will be above or below marginal cost, either because a reduction in the
number of users on one side has an infra-marginal impact on the surplus derived from the users of
the other side, or because the reduction in the number of users on one side may affect the
probability that a user on the other side finds a suitable partner with whom to transact. See
Hermalin, B.E, and M. L. Katz (2017) “What’s so special about two-sided markets?,” forthcoming
in Economic theory and public policies: Joseph Stiglitz and the teaching of economics, Columbia
University Press.
3 See, e.g., Wright, J., (2004) “One-sided logic in two-sided markets,” Review of Network
Economics, 3(1): 44-64, and Evans, D.S., and R. Schmalensee (2007) “The industrial organization
of markets with two-sided platforms,” Competition Policy International, 3(1): 151.
4 Evans, D.S. (2003) “The anti-trust economics of multi-sided platform markets,” Yale Journal on
Regulation, 20(2): 325-382.
5 That is, if users on one side of one the merging platforms can transact with users on the other side
of the other merging platforms.
6 See Chandra, A., and A. Collard-Wexler (2009) “Mergers in two-sided markets: an application to
the Canadian newspaper industry,” Journal of Economics and Management Strategy, 18(1), 1045-
1070; Fillistrucchi, L., Klein, T., and T. Michielsen (2012) “Assessing unilateral merger effects in
a two-sided market: an application to the dutch daily newspaper market”, Journal of Competition
Law and Economics, 8(2): 297-329; and and Baranes, E., Cortade, T. and A. Cosnita-Langlais
(2016), “Merger control on two-sided markets: is there need for an efficiency defense?,” NET
Institute, available at www.NETinst.org.
7 See Chandra, A., and A. Collard-Wexler (2009), op. cit. note 6.
8 See Jeziorski P. (2014) “Effects of mergers in two-sided markets: the U.S. radio industry,”
American Economic Journal: Microeconomics, 6(4): 35-73.
9 See Cantillon, E., and P.L. Yin (2011), “Competition between exchanges: a research agenda,”
International Journal of Industrial Organization,” 29 (3): 329-336.
10 See Pagano, M., and J. Padilla (2005a), “The economics of cash trading: an overview”, report
prepared for Euronext, and references therein.
11 See Arnold, T., P. Hersch, J.H. Mulherin, and J. Netter (1999) “Merging Markets”, Journal of
Finance, 54(3): 1083-1107.
12 Pagano, M., and J. Padilla (2005b), “Efficiency gains from the integration of exchanges: lessons
from the Euronext “natural experiment,” report prepared for Euronext.
13 Nielsson, Ulf (2009) “Stock exchange merger and liquidity: the case of Euronext,” Journal of
Financial Markets, 12(2): 229-267.
14 Data includes all securities that have composed each index at any time throughout the sample
period.
15 Annex A provides descriptive statistics on the bid-ask spreads provided by Bloomberg. Note that
these are normalised bid-ask spreads.
16 These techniques make it possible to estimate the relationship between two variables when the
variable under investigation is potentially influenced by many other factors.
17 A list of these events is reported in Annex B.
18 See Annex C for further description of the variables and data used in this paper.
19 See Euronext (2007), “Further information following the info-flash of 5 January 2007 regarding
new trading hours and tick size on the Euronext Cash Market.
20 We include the daily volume of securities listed in the CAC 40, BEL 20, AEX and PSI indices
traded on Chi-X and Bats.
21 See also Nielsson (2009, page 12) for further discussion of this issue.
22 The main econometric results in this paper are included in Tables 3 to 5 and can be found in
Annex D.
23 Results are unchanged if we replace this volatility measure for the volatility of the FTSE index.
See Annex E.
24 Multicollinearity occurs when some of the control variables are linearly related. When this linear
relationship is strong, the variation in the explanatory variables is insufficient to accurately
calculate the effect of these explanatory variables on the dependent variable. See Kennedy P.
(2003), A Guide to Econometrics 5th edition, Cambridge: MIT Press, Chapter 11.
25 This could be consistent with Pagano (1989) and Chowdry and Nanda (1991) who argue that
liquidity will tend to concentrate in a few markets if transaction costs are limited. See Pagano, M.
(1989), “Trading volume and asset liquidity”, Quarterly Journal of Economics, 104(2), 255-274
and Chowdry, B. and V. Nanda (1991), “Multimarket trading and market liquidity”, Review of
Financial Studies, 4(3), 483-511.
This paper examines the relationships among parties in a multi-sided market and
discusses how those relationships should affect the analysis of competitive effects and
efficiencies when competition agencies review conduct or transactions by multi-sided
platforms.
1
Howard Shelanski is a Partner at Davis Polk & Wardwell LLP and a Professor of Law at
Georgetown University. Samantha Knox and Arif Dhilla are associates at Davis Polk & Wardwell,
LLP. We thank Ilana Rice for assistance with this draft paper.
A positive network effect occurs when “the value that a customer on one side realizes
from the platform increases with the number of customers on the other side.” 5 For
example, eBay—through which individuals can buy and sell goods on line—becomes
more valuable to buyers as the number of sellers increases because there are more items
available for sale. At the same time, eBay becomes more valuable to sellers as the number
of buyers increases because there are more potential customers available. Network effects
need not be symmetric or even run in the same direction between two sides of a market.
Advertisers probably benefit from an increase in users more than users benefit from an
increase in advertisers, and in some cases users may even suffer detriment from increased
advertising. 6
Because network effects create interdependencies among the groups on a multi-sided
platform, a feedback loop may develop when membership of one side of the platform
grows or shrinks. To illustrate, assume a platform raises the price of platform access for
suppliers of some good or service. If some of those suppliers leave, the platform becomes
less valuable to customers on the other side of the market, who in turn also leave, further
reducing the platform’s value to the remaining suppliers, and so forth. 7 These dynamics
need not be perpetual or irreversible, but at some point they can go far enough to tip a
platform market toward failure or dominance. As discussed below, the effects of this
feedback loop may have important implications for both conduct and merger analyses.
Supply of such content or services might come from third parties (e.g., journalists and
musicians), the platform itself (e.g., Amazon and Netflix), or other platform users (e.g.,
Facebook, Twitter, Snapchat).
2. Assume Uber maintains fares but increases the percentage of the fare it keeps for
the company. That conduct appears directly to harm drivers and leave riders
unaffected. If this action causes drivers to leave the platform (or reduce the
number of rides they offer), however, it will also harm riders.
3. Assume Uber prohibits drivers from also driving for competing ride hailing
services. That policy might harm drivers while appearing to leave riders
unaffected. But if drivers leave the platform in response, the action will also harm
riders.
4. Assume Uber prohibits riders from riding with competing services. The action
directly harms riders but not drivers, unless riders abandon Uber in response.
While it is difficult to predict the extent of the benefit or harm caused, regulators should
be aware that conduct harming one side of a service-based relationship has the potential
to result in harm to the other side (and vice versa). Depending on conditions and indirect
effects, conduct that at first look appears to affect parties differently may have effects that
are positively correlated across different sides of the market, thereby exacerbating either
the harms or the benefits.
Identifying the relationships across a platform as either service-based or subsidy-based
can therefore be important to predicting the relative balance of network effects among the
different sides of the market. As discussed below, the balance and direction of network
effects can have important implications for how regulators should evaluate net welfare
effects of conduct and transactions by multi-sided platforms.
should consider each of the efficiencies in determining whether the action was
anticompetitive.
Workflow for conduct analyses by competition agencies. When analysing the effect of
challenged conduct on a multi-sided market, regulators should first look at the
relationship between the targeted side of the platform and the other sides of the
platform. 20 When conduct targets a side of the market participating in a service-based
relationship, regulators should closely evaluate the potential impacts on all sides of the
market given the likely cross-platform network effects, regardless of which side the
conduct targets. If, in contrast, a platform directs its conduct toward subsidisers, cross-
platform effects are much less likely unless the action would drop subsidies to insufficient
levels, which would render the conduct economically senseless and therefore unlikely to
continue or to occur in the first place.
With respect to efficiencies, therefore, agencies need to take into account how
efficiencies that flow directly to users or suppliers will also indirectly affect other sides of
the market through cross-network effects. When the efficiencies directly benefit
subsidisers, there is less likelihood of such cross-network effects benefitting users or
suppliers. The implication is that if conduct is efficient for suppliers or users but might
raise prices for subsidisers, the reviewing agency should consider whether the efficiencies
offset that possible harm to subsidisers through cross-platform externalities. Where, on
the other hand, the conduct’s direct effect is to harm users or suppliers, it is less likely
that efficiency gains for the subsidisers will offset those harms, unless that benefit to the
subsidiser is necessary to attract or maintain necessary subsidy levels. It therefore may be
more important for authorities to consider efficiency effects on all sides of a multi-sided
platform when those efficiencies benefit users or suppliers than when they benefit
subsidisers.
model illustrated that an increase in subscription prices was likely to have a negative
effect on both subscriber demand (resulting in lower circulation) and on advertising
revenue (since decreased circulation leads to less demand for advertising). 24 The authors
concluded that “raising the newspaper price is likely to lead not only to a loss in readers
but also to a loss in advertising, and therefore the tendency to increase prices will be
lower than in the absence of network effects.” 25
Subsidisers may be more vulnerable to unilateral effects than service providers or
users. The dependence of the demand for advertising on the number of platform users
leads to a closely related corollary: subsidisers might be more vulnerable to the unilateral
effects of a merger than other platform participants. As illustrated above, strong network
effects can serve as an independent pricing constraint on a platform’s incentive and
ability to raise prices. Because the network effects in a subsidy-based relationship are
skewed heavily towards the subsidiser, however, the relatively weak network effects
experienced by users and content providers might not provide the same constraint on
price increases to advertisers. Indeed, several recent studies suggest that consolidation of
multi-sided platforms results in higher prices to subsidisers. 26 While more research is
needed to test this observation, this apparent effect may be explained by the fact that
platforms can drive user demand by increasing rates to advertisers and decreasing
subscription costs or increasing quantity or quality for users. 27
Platform mergers that result in price increases may yield net efficiencies. Even if it is
true that subsidisers would be subject to price increases following a merger, the merger
could nonetheless yield welfare gains—including for the subsidisers themselves. For
example, Song’s study finds that both the average surplus and the total surplus to
advertisers went up at magazines that increased advertising prices post-merger because
the lower copy prices raised the number of subscribers and in turn the audience for the
advertisers. 28 The study found that although “[a]dvertisers . . . usually face higher ad
prices in more concentrated markets . . . they are not necessarily worse off if lower copy
prices attract a large number of readers.” 29 This result does not imply that subsidisers will
always gain from a merger of multi-sided platforms, but it does imply that the efficiencies
analysis of such a merger should take into account the cross-platform externalities of any
merger-related increase in the number of users to whom advertisers will have access.
Presumably, subsidisers will always benefit from an increase in a platform’s
subscriber/user base. 30 A merger of overlapping multi-sided platforms will necessarily
result in such an increase. It is therefore plausible that any post-merger price increase to
subsidisers could be entirely offset by the increase in the subscriber/user base that results
from the merger. Cross-platform network effects must therefore inform the efficiencies
analysis in mergers of multi-sided markets.
3. Conclusion
This paper has examined how the nature of the relationships among the different sides of
a multi-sided platform can affect the direction, magnitude, and relative balance of cross-
platform network effects. Whether a platform’s actions affect parties engaged in subsidy
relationships or service relationships has important implications for evaluating the
competitive effects and efficiencies. For example, the extent to which efficiencies that
flow directly to one side of the market will have positive externalities that offset
competitive harm to another side of the market will depend in part on whether the direct
beneficiary is a consumer, supplier or subsidiser for the potentially harmed side of the
platform. Efficiency gains to users and suppliers are more likely, through the network
Notes
1 See Gönenç Gürkaynak et al., Multisided Markets and the Challenge of Incorporating Multisided
Considerations into Competition Law Analysis, 5 J. ANTITRUST ENFORCEMENT 100, 101–05
(2017) (describing various definitions of multisided markets).
2 The delegates at the 2009 OECD Policy Roundtable on Two-Sided Markets recognised that both
(1) distinct groups and (2) network effects between the groups were essential elements of a two-
sided market. Secretariat, Executive Summary, in POLICY ROUNDTABLES: TWO-SIDED MARKETS
11, 11 (Organisation for Economic Co-operation and Development Competition Committee,
2009). See also, e.g., Gürkaynak et al., supra note 1, at 101–05; Jan Ondrus et al., The Impact of
Openness on the Market Potential of Multisided Platforms: A Case Study of Mobile Payment
Platforms, 30 J. INFO. TECH. 260, 261 (2015); David Evans, Background Note, in POLICY
ROUNDTABLES: TWO-SIDED MARKETS 23, 29 (Organisation for Economic Co-operation and
Development Competition Committee, 2009); David S. Evans & Richard Schmalensee, The
Industrial Organization of Markets with Two-Sided Platforms, 3 COMPETITION POL’Y INT’L 151,
152 (2007); Renata B. Hesse & Joshua H. Soven, Defining Relevant Product Markets in
Electronic Payment Network Antitrust Cases, 73 ANTITRUST L.J. 709, 714 (2006). In contrast,
Andrei Hagiu and Julian Wright believe that “network effects are neither necessary nor sufficient”
in defining multisided platforms. Andrei Hagiu & Julian Wright, Multi-sided Platforms, 43 INT’L
J. INDUS. ORG. 162, 164 (2015).
3 Evans, supra note 2, at 29.
4 In the context of multisided markets, network effects have also been referred to as cross-group
externalities or indirect externalities. See, e.g., Mark Armstrong, Competition in Two-Sided
Markets, 37 RAND J. ECON. 668 (2006); Secretariat, supra note 2, at 11.
5 Evans, supra note 2, at 29. Network effects may be either direct or indirect. “Direct network
effects arise where users of the product interact with each other, so having more users makes the
product more useful and valuable.” Secretariat, Executive Summary, in THE DIGITAL ECONOMY 5,
8 (Organisation for Economic Co-operation and Development Competition Committee 2012).
6 See, e.g., Evans, supra note 2, at 24 (referring to situations where the sides exhibit unbalanced
network effects); Secretariat, supra note 2, at 12 (same).
7 See, e.g., Evans & Schmalensee, supra note 2, at 159 (describing a feedback loop created by
raising prices on one side of a platform).
8 Advertisers might convey information about available goods and services. See, e.g., Evans &
Schmalensee, supra note 2, at 155–156 n.10. Nonetheless, users do not join a platform seeking
advertising in the same way that they seek the platform’s primary content or services (as shown by
the fact that on some subsidy platforms, users opt to pay to avoid ads (e.g., Pandora Premium).
9 See David S. Evans & Richard Schmalensee, The Antitrust Analysis of Multisided Platform
Businesses, in THE OXFORD HANDBOOK OF INTERNATIONAL ANTITRUST ECONOMICS 404, 410
(Blair & Sokol eds., 2015) (observing that most advertiser-supported media combine content with
advertising to attract consumers).
10 See Evans, supra note 2, at 24 (referring to situations where the sides exhibit unbalanced network
effects); Secretariat, supra note 2, at 12 (same).
11 Cf. Evans & Schmalensee, supra note 9, at 429 n.29 (noting that studies of the newspaper industry
suggest that advertising does not produce positive or negative externalities for readers, but other
forms of advertising may be valued by consumers).
12 Uber is an online platform that connects people seeking car rides (“riders”) with people who are
willing and available to offer rides (“drivers”). Uber sets the fare for a ride and then charges the
driver a percentage of each trip’s fare as a fee.
13 See, e.g., U.S. Dep’t of Justice & Fed. Trade Comm’n, Horizontal Merger Guidelines §10 (2010)
[hereinafter 2010 Horizontal Merger Guidelines]. While merging parties have raised affirmative
defenses based on efficiencies, no such defense has succeeded in saving a transaction that was
otherwise found by the court to be anticompetitive. See, e.g., FTC v. Penn State Hershey Med.
Ctr., 838 F.3d 327, 347-48 (3d Cir. 2016); FTC v. H.J. Heinz Co., 246 F.3d 708, 720-22 (D.C.
Cir. 2001); FTC v. University Health, 938 F.2d 1206, 1222-24 (11th Cir. 1991).
14 See United States v. Microsoft Corp., 253 F.3d 34, 59 (D.C. Cir. 2001) (stating that greater
efficiency can constitute a defendant’s “procompetitive justification for its conduct”).
15 2010 Horizontal Merger Guidelines, § 10 (“[I]t is incumbent upon the merging firms to
substantiate efficiency claims so that the Agencies can verify by reasonable means the likelihood
and magnitude of each asserted efficiency, how and when each would be achieved (and any costs
of doing so), how each would enhance the merged firm’s ability and incentive to compete, and
why each would be merger-specific”).
16 See, e.g., 2010 Horizontal Merger Guidelines § 10 (“Cognizable efficiencies are merger-specific
efficiencies that have been verified and do not rise from anticompetitive reductions in output or
service.”); 3 PHILLIP E. AREEDA & HERBERT HOVENKAMP, ANTITRUST LAW: AN ANALYSIS OF
ANTITRUST PRINCIPLES AND THEIR APPLICATION ¶ 658f (4th ed. 2015) (“Thus when courts speak
of the business justification defense as requiring some showing of ‘efficiency,’ that term should be
understood to refer to the costs or output of the monopolist itself (productive efficiency), not to
the market as a whole (allocative efficiency).”).
17 AREEDA & HOVENKAMP, supra note 13, at ¶ 658f; 2010 Horizontal Merger Guidelines § 10
(efficiencies that “reduce the incremental cost of production” are more likely to be cognizable); In
both the merger and the unilateral conduct contexts, courts and agencies appear to require that
efficiencies increase consumer welfare in order to be cognisable. See, e.g., 2010 Horizontal
Merger Guidelines § 10; Data Gen. Corp. v. Grumman Sys. Support Corp., 36 F.3d 1147, 1183
(1st Cir. 1994) (“In general, a business justification is valid if it relates directly or indirectly to the
enhancement of consumer welfare.”).
18 United States v. Am. Express Co., 838 F.3d 179, 197-98, 202-05 (2d Cir. 2016) (holding that what
mattered was the adverse effect on competition “as a whole” and that the whole market included
both sides); Case C-67/13 P, Groupement des Cartes Bancaires (CB) v. Comm’n, 2014 E.C.R.
2204.
19 See, e.g., Lapo Filistrucchi et al., Assessing Unilateral Effects in a Two-Sided Market: An
Application to the Dutch Daily Newspaper Market, 8 J. COMP. L. & ECON. 297, 301 (2012)
(providing example of how a newspaper that raises the subscription price should account for the
“negative effect on advertising revenues as decreased circulation leads to a decline in the demand
for advertising”).
20 If the multisided market has more than two sides, it is possible that the side is involved in multiple
relationships. For example, if Pandora, an online radio station, takes an action against users, the
users will be in both (1) a subsidy-based relationship with the advertisers, and (2) a service-based
relationship with the musicians providing content.
21 Evans & Schmalensee, supra note 9, at 428 (“[A]ll else equal a merger of multisided platforms
would ordinarily increase indirect network externalities by increasing the size of all customer
groups and thereby provide efficiency benefits”).
22 See 2010 Horizontal Merger Guidelines § 1 (“Enhancement of market power by sellers often
elevates the prices charged to customers.”).
23 Minjae Song, Estimating Platform Market Power in Two-Sided Markets with an Application to
Magazine Advertising 30–33 (Simon School, Working Paper No. FR 11-22, 2013), available at
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1908621.
24 Filistrucchi et al., supra note 22, at 325-26.
25 Id. at 326.
26 See, e.g., id.; Song, supra note 29; Przemyslaw Jeziorski, Effects of Mergers in Two-Sided
Markets: Examination of the U.S. Radio Industry, 6 AM. ECON. J. MICROECONOMICS 35, 36
(2014) (empirical study of consolidation in U.S. radio industry from 1996 to 2006 found merger
wave resulted in 6% increase in prices to advertisers).
27 See Song, supra note 29.
28 Id. at 32.
29 Id. at 33.
30 See id.
By Paul A. Johnson 1
This note discusses some key questions that investigations should consider when
assessing vertical restraints in multi-sided platforms. It is composed of three main
sections. The first formulates a threshold question: when should we apply the economics
of platforms to an analysis of vertical restraints? The last two sections, assuming the
previous question has been answered affirmatively, address the economic assessment of
anticompetitive and procompetitive effects of vertical restraints in platforms.
network a “merchant price,” for every merchant price it receives, the platform also
receives a “consumer price” (which can be a negative price indicating that consumers are
paid to use the card). By providing the transaction service, the platform retains the sum of
the two prices, a property first recognised by then-US Department of Justice Assistant
Attorney General Baxter. 2 Correctly defining the notion of “price” is critical when
assessing whether a vertical restraint allows a payment card platform to expand or
preserve its market power (i.e., charge a supracompetitive price).
Other examples lie between that of an “automobile platform” and a ride-sharing or
payment card platform. For example, a newspaper can change the number of
advertisements it sells without the number of subscribers automatically changing, and
vice-versa. The Google search engine is similar in that advertisements do not always
appear after a search. Moreover, the Google platform is broad in that it offers a number of
other services to consumers beyond search. Some of these services, like Google Scholar,
never appear to show advertising. Thus, Google can change the number of advertisements
it shows to users without users changing the intensity of their engagement with the
platform. Amazon is similar. While every transaction on Amazon Marketplace involves
exactly one consumer and one third-party merchant, Amazon earns revenues from
multiple products (e.g., retail offerings offered by Amazon, streaming video, an appstore,
tablets). Amazon could draw more users to its platforms with these other products and
thereby increase third-party merchant sales on Amazon marketplace. However, such an
increase is not an automatic and necessary result of increased user participation on the
platform.
The difference in these examples can be appreciated without resorting to sophisticated or
nuanced economic reasoning. Instead, the difference is driven by the degree to which
quantity transacted on one side changes with quantity transacted on the other side; “fixity
of use” will refer to this degree. It differs from network effects, usually defined with
reference to benefits of membership increasing as other users join, in that fixity of use
emphasises use. 3 The importance of fixity of use may not be always appreciated. For
example, some maintain that a “mature” payment card network is not two-sided based on
a claim that the magnitude of network effects diminishes as a platform “matures” (i.e., as
network effects diminish). 4 That conclusion incorrectly ignores a platform’s need to
encourage use by its members independent of its “maturity.” And, at a more basic level,
that conclusion incorrectly ignores the fixed-proportions nature of a payment card
platform and does not recognise the price the platform receives from facilitating a
transaction.
To appreciate fixity of use further, recognise the difference between use of a service
offered by a platform and membership on a platform. For example, riders and drivers can
choose to become members of a ride-sharing platform after which they choose how
intensively to use that platform. In that sense, there is not necessarily “fixity of
membership” in a ride-sharing platform: the number of riders that have an Uber account
can change without the number of drivers changing. Similarly, the number of merchants
that accept a particular payment card can change independent of the number of consumers
who hold the card. Depending on whether an investigation centers on transaction-specific
fees or on the fees paid by users that are independent of usage, that distinction is
important to recognise. 5 For example, suppose that it is determined that merchant
acceptance of a payment card is not affected by a small change in the number of
consumers who hold the card. In that case, an analysis of any membership fees merchants
pay to accept the card need not consider annual fees paid by consumers. That conclusion,
however, does not change the fact that every usage of the payment platform necessarily
implicates exactly one merchant and exactly one consumer and that the payment platform
retains the sum of the merchant and consumer prices.
The degree of fixity of use should be determinative of whether a platform’s services and
prices are assessed akin to a payment platform (i.e., where multi-sided principles are
important) or akin to an automobile “platform” (i.e., where multi-sided principles can be
ignored). 6 As a simple illustration, suppose a platform seeks to improve the match
between two sets of users, who are charged very different prices and offered very
different services. As a first step, one might consider whether it is meaningful to consider
the terms offered to one set of users independent of the terms offered to the other set of
users. The degree of fixity of use should be determinant because the analysis should
reflect the platform’s business realities: as fixity of use disappears, the platform treats the
two sides independently. 7 In other words, if the two sides operate reasonably
independently, then an analysis can reliably analyse either side independent of the others.
However, when the operations of both sides of the platform are reasonably tightly linked,
a reliable analysis should consider both sides jointly. That statement operates
independently of, for example, how antitrust markets are defined or how strong network
effects are because it speaks to the price that is relevant to the platform. The overall price
relevant to the platform can be calculated as the quantity-weighted average of prices paid
by each side. 8
Another important implication of economic logic is that a vertical restraint cannot be the
source of bargaining power but is the result of bargaining power. That implication leads
to the conclusion that a vertical restraint that limits one entity’s actions is accompanied by
prices, or other terms, that compensate for that limitation. 13 For example, if a vertical
restraint moves some risk from a buyer to a seller, then the price paid to the seller must
increase in compensation. The inclusion of such a vertical restraint cannot somehow
endow the buyer with bargaining power because such a claim is circular (i.e., the restraint
would have to create the bargaining power that was necessary for its imposition it in the
first place).
The logic of the previous two paragraphs is not always explicitly recognised in economic
models that study vertical restraints. That is not to say that the logic is not widely
accepted—it is—rather, the omission reflects a deliberate choice to simplify and focus
economic models on a small set of issues. For example, a model may analyse the effects
of a vertical restraint by comparing outcomes with and without the vertical restraint while
holding all else constant and imposing some restriction on the form that contracts can
take (e.g., linear pricing). This analysis implicitly compares two very different settings
(specifically, two very different extensive form games) where the relative bargaining
power of the parties differs substantially. In richer settings, firms will react to a ban on
the vertical restraint by changing aspects of their behavior that the model’s restrictive
assumptions do not permit. For example, if an entity has power and the ability to impose
a restraint, it also has the option to exercise that power differently—say simply by
charging a higher price. While economic models may abstract away from such important
issues relating to the existence of a vertical restraint, that abstraction does not allow
enforcement authorities to do the same.
An interesting application of the logic of vertical restraints is evident in the United States
Department of Justice (USDOJ) complaint against Blue Cross Blue Shield of Michigan
(BCBSM). 14 BCBSM might be viewed as a platform because it serves two separate sets
of entities: hospitals and payers of health insurance. USDOJ alleged that BCBSM signed
contracts that included most favored pricing terms with certain hospitals in Michigan,
which committed the hospitals to sell BCBSM medical services at prices that were lower
than any other entity received. USDOJ alleged that the effect of these terms was to limit
competition in certain markets by limiting the competition that BCBSM faced. Notably,
the complaint alleged that BCBSM obtained most favored pricing terms from some
hospitals “by agreeing to increase its payments to the hospital.” 15 The victims, thus, were
not the hospitals who signed contracts with the vertical restraint—they received higher
payments—but were third party payers of health insurance and competing health
insurance providers.
This discussion implies that not only should competition authorities have a strong
presumption that vertical restraints are bilaterally efficient (although not necessarily
socially efficient), but that they should be cognisant that the complaints they hear may be
driven by pre-existing market power rather than a legitimate harm to competition.
Specifically, parties who have signed a vertical restraint may not be “happy” with the
price or services they receive even if the vertical restraint is procompetitive. Ultimately, a
complaint may simply reflect a perception of a lack of bargaining power and be unrelated
to any anticompetitive effects of the vertical restraint. This danger is most acute when the
complainants are managers whose job responsibilities are narrowly circumscribed to
achieve as low (or high) a price as possible without having a broader understanding of
their firm’s operations.
NSR as a means for a payment network to “tax” rival payments methods. 20) In general,
the effects on consumer welfare are ambiguous. The merchant’s market power distorts the
credit price away from optimal levels and an NSR mitigates this effect; however, the NSR
introduces additional distortions (namely an increase in the cash price as well as a
distortion in the fees charged by the payment card network) that can outweigh its
efficiency enhancing effects. In the Schwartz and Vincent model an NSR can either
increase or decrease consumer surplus and total surplus.
How are the restraint’s effects transmitted to other sides of the platform?
In assessing a vertical restraint, some jurisdictions balance any anticompetitive effects
against any procompetitive effects. When a platform is not at issue, a single set of
consumers feels the effects of those opposite effects. In a platform, however, those
opposite effects may be felt by distinct sets of users. This dynamic suggests that
competition authorities should analyse whether and how effects of vertical restraints are
transmitted across different sides.
As an example, consider the United States Department of Justice’s (USDOJ) case against
BCBSM, which involved most favoured pricing terms and was described in section 0. .
One approach might be to weigh the procompetitive and anticompetitive effects of the
restraints considering only the relationships between hospitals and insurance companies.
According to one such anticompetitive theory, those hospitals that had not incorporated
the vertical restraint into their contracts would suffer from an elimination of competition
among insurance companies; those hospitals that had incorporated the vertical restraint
into their contracts could benefit. USDOJ, however, noted that higher prices paid to
hospitals by BCBSM likely result in higher prices paid by a different set of users: payers
of health insurance. Specifically, USDOJ noted that one set of payers, known as “self-
funded” payers pay for their own healthcare claims. 21 Such payers would be harmed
because they pay more when hospital prices increase. USDOJ also noted that state law
allowed BCBSM to base the premiums of “fully insured” payers on historical healthcare
costs “so increases in local hospital prices can lead directly to increased premiums.” 22
More generally, a vertical restraint on one side of a platform, by definition, restrains an
expression of competition on that side. However, it can also have effects on other sides of
the platform. In principle, those effects can benefit or harm the other side. USDOJ’s
complaint against BCBSM is an example where a vertical restraint on one side harms
users on another side. Another theory might hold that a vertical restraint limits
competition on one side but ultimately benefits users on another side due to a shift in the
locus of competitive vigour. (In this context, it is useful to keep in mind that a vertical
restraint must be bilaterally efficient before parties accept it, so it is necessary to explain
why users on one side are better off by agreeing to a vertical restraint that limits one
expression of competition on their side.)
The term “waterbed effect” has been used in the telecommunications literature to describe
the effect of fixed-to-mobile termination rates on the prices paid by mobile telephone
users. 23 Empirical analysis in that literature has exploited shifts in regulation over time to
estimate how different sets of users are affected. In principle, such an approach could be
used more generally to estimate how effects of vertical restraints are transmitted across
platforms. 24
While vertical restraints restrain, they can also enable expressions of competition. Thus,
an analysis that focuses exclusively on what a restraint prevents without considering what
it enables is incomplete. This section discusses two common procompetitive justifications
for vertical restraints in platforms.
• Can free riding be resolved in other ways? Theory allows for multiple ways to
solve free riding. For example, the platform could discontinue to serve offending
users. Alternatively, a platform might be able to use a fee structure that involves a
lump-sum payment that eliminates any incentive to free ride. Asking whether a
vertical restraint is the “least restrictive alternative” is, thus, not possible to
answer generally, but an important question to answer within a specific context.
Notes
1 Marc Rysman. “The economics of two-sided markets.” The Journal of Economic Perspectives 23,
no. 3 (2009): 125-143, 127.
2 William F. Baxter. “Bank interchange of transactional paper: Legal and economic perspectives.”
The Journal of Law & Economics 26, no. 3 (1983): 541-588, 545. (“Perhaps the most intuitively
appealing way to resolve the difficulties posed by this market model is to redefine what we mean
as one unit of the product consumed. Rather than considering the demands of [the purchaser] P
and [the merchant] M as demands for separate products, define one unit of product to consist of
the bundle of transactional services that banks must supply jointly to P and M in order to facilitate
the execution of one exchange of goods or services between P and M. Under this interpretation,
the supply price of the product is the sum of the individual charges to P and to M. Furthermore,
the demand for that product is a joint demand of P and of M: in combination they must make a
payment of that magnitude to the banks to induce the necessary supply, but independently neither
P nor M necessarily confronts any particular price as one he must pay in order to have his demand
fulfilled.”)
3 Michael L. Katz and Carl Shapiro. “Systems competition and network effects.” The Journal of
Economic Perspectives 8, no. 2 (1994): 93-115, 94. (“Because the value of membership to one
user is positively affected when another user joins and enlarges the network, such markets are said
to exhibit ‘network effects,’ or ‘network externalities.’”)
4 For example, see Alan S. Frankel and Allan L. Shampine. “The economic effects of interchange
fees.” Antitrust Law Journal 73, no. 3 (2006): 627-673, 655. (“By its nature, a network externality
is likely to become less important . . . as a network matures.”).
5 For example, some credit cards charge consumers an annual fee that is independent of usage. In
the United States, Visa charges some merchants a “fixed acquirer network fee” that may be
independent of usage.
6 Jean-Charles Rochet and Jean Tirole have defined a platform to be “one-sided” if the volume of
transaction depends only on the aggregate price level and not on the structure. But because their
focus appears to be mainly on payment platforms and other platforms where fixity of use is
perfect (e.g., bilateral electricity trading), their definition identifies instances when parties can
negotiate bilaterally to “undo” any particular price structure and not on the extent to which
quantity on one side could increase independent of quantity on other sides. See Jean‐Charles
Rochet and Jean Tirole. “Two‐sided markets: a progress report.” The RAND Journal of Economics
37, no. 3 (2006): 645-667, 648.
7 To be more specific, consider a platform with two sides, A and B. To simplify, suppose that use
by side B is relevant to side A, but use by side A is not relevant to side B. Costs to serving either
side are zero. In this case the profit of the platform can be written as 𝜋𝜋 = 𝑃𝑃𝐴𝐴 𝑄𝑄𝐴𝐴 �𝑃𝑃𝐴𝐴 , 𝑄𝑄𝐵𝐵 (𝑃𝑃𝐵𝐵 )� +
𝜕𝜕𝜕𝜕
𝑃𝑃𝐵𝐵 𝑄𝑄𝐵𝐵 (𝑃𝑃𝐵𝐵 ). The first-order condition of profit with respect to the side-B price is =
𝜕𝜕𝑃𝑃𝐵𝐵
𝜕𝜕𝑄𝑄𝐴𝐴 𝜕𝜕𝑄𝑄𝐴𝐴
𝑃𝑃𝐴𝐴 𝑄𝑄𝐵𝐵′ (𝑃𝑃𝐵𝐵 ) + 𝑄𝑄𝐵𝐵 + 𝑃𝑃𝐵𝐵 𝑄𝑄𝐵𝐵′ (𝑃𝑃𝐵𝐵 ) = 0. The term reflects fixity of use. The first term in the
𝜕𝜕𝑄𝑄𝐵𝐵 𝜕𝜕𝑄𝑄𝐵𝐵
first-order condition approaches zero as fixity of use disappears leaving only the latter two terms.
The latter two terms represent the standard first-order condition if side B were independent of A.
8 For example, in a two-sided platform with sides A and B, the quantity-weighted average of the
𝑄𝑄 𝑄𝑄
prices paid by each side is 𝑃𝑃 = 𝑃𝑃𝐴𝐴 𝐴𝐴 + 𝑃𝑃𝐵𝐵 𝐵𝐵 .
𝑄𝑄𝐴𝐴 +𝑄𝑄𝐵𝐵 𝑄𝑄𝐴𝐴 +𝑄𝑄𝐵𝐵
In some cases, however, the price to one side may be set to zero and platforms compete by
providing services to attract users. In theory, a platform’s profits are affected in the same way
whether it spends a dollar on providing a service or it spends a dollar lowering the price. In
practice, quantifying the costs of providing these non-price attractions may be challenging.
9 An excellent reference is Chapter 4 of Michael D. Whinston. “Lectures on Antitrust Economics.”
MIT Press Books 1 (2008).
10 The proof is easy and instructive. A buyer and seller are negotiating to trade an item. The buyer
values the item at v dollars; the seller values the item at c dollars. Suppose gains to trade exist
(i.e., v-c>0), but the parties do not trade. Either the buyer or seller could propose a price between
c and v that, if accepted, would leave both sides strictly better off. Doing so is possible because c
and v are common knowledge; doing so is also nearly costless because bargaining costs are small.
11 First suppose that the clause is in the contract but it is inefficient. The buyer and seller can both be
made better off if the clause is taken out, so they will take it out. Next suppose that the clause is
efficient but is not included in the contract. Again, by including the clause, each party can be
made better off by including it, so they will include it.
12 Whinston, supra note 9 at 140, writes “In recent years, a number of authors have shown how
sensible alterations to this Chicago School model can make exclusive contracts a profitable
strategy for excluding rivals. These models all have the feature that some form of externality
arises from an exclusive contract signed by two parties onto other individuals, and this externality
makes the contract jointly optimal for the contracting parties.”
13 An exception involves an incumbent seller who can make sequential offers to a large number of
buyers to exclude potential rival sellers. In that setting, the compensation to the buyers for signing
an exclusive contract approaches zero. In this way, signatories to vertical restraints can be harmed
not by their own restraints (which do not harm them but offer little compensation), but through the
agreements others sign that foreclose entry. See Eric B. Rasmusen, J. Mark Ramseyer, and John S.
Wiley Jr. “Naked exclusion.” The American Economic Review (1991): 1137-1145 and Ilya R.
Segal and Michael D. Whinston. “Naked exclusion: comment.” The American Economic Review
90, no. 1 (2000): 296-309.
14 Complaint, United States and State of Michigan v. Blue Cross Blue Shield of Michigan. Case
2:10-cv-14155-DPH–MKM. Available at https://www.justice.gov/atr/case-
document/file/489536/download.
15 Id. ¶ 44. See also ¶¶ 40, 49, 58, 60, 68, 75.
16 Eastman Kodak Co. v. Image Technical Services, Inc., 504 US 451 (1992).
17 Carl Shapiro. “Aftermarkets and consumer welfare: Making sense of Kodak.” Antitrust Law
Journal 63, no. 2 (1995): 483-511, 497. (“I am not convinced that this type of consumer injury is
worthy of the attention of antitrust laws.”)
Setting aside whether such a distortion ought to be worthy of attention from an antitrust authority,
much of the literature has concluded that market power may create distortions of ambiguous sign
on how platforms allocate the price among different sides. In other words, while market power
may introduce a distortion in how the total price is allocated among the sides, that distortion does
not necessarily favor one side or another. Glen Weyl analogises this distortion to the more familiar
result, due to Michael Spence, that a monopolist may either over- or under-provide quality
depending on the nature of heterogeneous consumer tastes for quality. E. Glen Weyl. “A price
theory of multi-sided platforms.” The American Economic Review 100, no. 4 (2010): 1642-1672.
18 Marius Schwartz and Daniel R. Vincent. “The no surcharge rule and card user rebates: Vertical
control by a payment network.” Review of Network Economics 5, no. 1 (2006).
19 Suppose that a merchant sells two goods that are independent in the sense that the price of one
good does not affect demand for the other good. Furthermore, suppose that, without any
constraint, the merchant would choose to set a higher price for good 1 than for good 2. If the
manufacturer of good 1 requires that the merchant’s selling price of good 1 not exceed the selling
price of good 2 then, making some additional regularity assumptions, the merchant’s constrained
(single) profit maximising price for the two goods will fall between the unconstrained prices of
the two goods. The merchant will raise the price of good 2 to some degree because it is required to
do so in order to raise the price of good 1 closer to its preferred, unconstrained price.
20 Joseph Farrell. “Efficiency and competition between payment instruments.” Review of Network
Economics 5, no. 1 (2006).
21 Supra, note 14, ¶ 15.
22 Supra, note 14, ¶ 18.
23 Christos Genakos and Tommaso Valletti. “Testing the ‘waterbed’ effect in mobile telephony.”
Journal of the European Economic Association 9, no. 6 (2011): 1114-1142.
24 See Marc Rysman and Julian Wright. “The economics of payment cards.” Review of Network
Economics 13, no. 3 (2014): 303-353. section 5.2 for a discussion of some potential difficulties of
estimating waterbed-type effects in the case of payment cards.
25 Jean‐Charles Rochet and Jean Tirole. “Must‐Take Cards: Merchant Discounts and Avoided
Costs.” Journal of the European Economic Association 9, no. 3 (2011): 462-495, 467.
The notation ps>bs reflects the situation when the merchant’s private benefits of accepting the
card are lower than the price it pays to do so. Setting ps=bs reflects the “Baxter fee.” Importantly,
a merchant may have an incentive to accept cards even when the price exceeds the benefits due to
a “business stealing” effect.
26 Michael L. Katz and Carl Shapiro. “Systems competition and network effects.” The Journal of
Economic Perspectives 8, no. 2 (1994): 93-115.
27 Robin S. Lee. “Vertical integration and exclusivity in platform and two-sided markets.” The
American Economic Review 103, no. 7 (2013): 2960-3000.
28 One of the most profound implications of network effects for antitrust is the implied tradeoff
between quality and competition. Specifically, competition may grow with the number of
competitors, but the benefits of network effects may be lost as the industry becomes fragmented.
Lee argues that consumer welfare would have likely increased substantially with concentration in
video game platforms. Marc Rysman’s study of yellow pages found the opposite: despite the
presence of network effects, more competition increased welfare. Marc Rysman. “Competition
between networks: A study of the market for yellow pages.” The Review of Economic Studies 71,
no. 2 (2004): 483-512.
1. Introduction
1
Paper submitted by Cristina Caffarran, Charles River Associates, and Kai-uwe Kühn, University
of East Anglia, Cepr and Dice
redundant and displace more productive ones. As a result, online distribution of goods
and services in Europe is being held back and in danger of ending up a lot less varied and
efficient than it should be.
After a brief introduction to multi-sidedness and the questions we set out to consider
(Section 2), in Section 3 we explain how the key insight that vertical restraints are
motivated by contractual incompleteness carries over directly to multi-sided markets.
Similarly in Section 4 we argue that while multi-sidedness may appear to complicate
greatly the analysis of competitive effects, the assessment can be simplified to an
approach close to the approach to vertical restraints in more standard environments.
Where a new approach is truly needed is in developing and taking seriously the evidence
for the efficiency properties of these restraints, which tend to be systematically
overlooked or dismissed (Section 5). Compounding the current problem is the structure of
the law, which involves a sequential assessment of anticompetitive effects and
efficiencies and is particularly ill-suited to vertical restraints; and possibly the reality that
Industrial Organisation research continues to put too much emphasis on the details of
specific models and is not encouraging enforcers to look beyond analyses of short-term
price effects, at dynamic issues of investment and experimentation (Section 6). Building
on these considerations, in Section 7 we make a number of suggestions for a roadmap to
the analysis of these practices. Section 8 concludes.
Markets are described as “multi-sided” when they are organised around an intermediary
(a “platform”) with interdependencies in demand between agents performing and
obtaining services on various sides of the platform. While there are multiple classic
examples (TV and newspapers, payment cards), for purposes of this paper we focus on
digital platforms that connect different constituencies of users: consumers searching for
information and a product/service to “match” their requirements; sellers looking to realise
a sale; advertisers serving up adverts to match and anticipate users’ interests; and the
platform itself, looking to monetise its services (information, matching) through
advertising and various other sales commissions.
Multi-sided markets involve a number of characteristics: (a) there are typically network
externalities across the sides of the platform; (2) the platform has incentives to invest to
develop a user base as wide as possible on one side, so that it can monetise its investment
on the other side (e.g. through advertising revenues on the other side, and as well as
through commission on sales/bookings); (3) this typically involves offering an attractive
service “for free” (or at a low price) to build up customer base on one side quickly; and
(4) investments in functionalities which are provided to users for “free” are susceptible to
free riding if they are available to all, but there is a separate channel through which
purchases/bookings can be made.
The question we discuss in this paper is whether we need to make changes to the
competition analysis of vertical contracts in these settings. For instance, contracts that
introduce restrictions on the prices that can be charged by sellers across channels (e.g.
“best price clauses”); on the distribution channels that may be used (e.g. brands allowing
distributors to use certain online marketplaces but not others); and on branding and
features that may be displayed (e.g. prohibition to use a logo on a platform). Do we need
new insights and new tools to deal appropriately with these cases?
A broad insight provided by economic analysis over the past 30-40 years on the
motivation for vertical restraints is that these give rise to anticompetitive effects only in
limited, very specific circumstances.2 Vertical co-ordination does not typically lower the
competitive pressure faced by a firm, but allows it to organise sales in a more effective
way. If a firm “restricts” its own downstream distribution, it does not affect directly its
competitors but restricts access to market of its own goods. As brand owners have an
interest in the distribution of their products being as competitive as possible, the question
is why would they limit the channels of distribution they use, or leave “money on the
table” in the form of a greater margin to the retailer? The most natural explanation in
most cases is that they can only be interested in doing so if this creates incentives for
beneficial activities that cannot be otherwise directly mandated and controlled. While
there is a literature on the anticompetitive effects of certain vertical restraints (e.g. the
classic case of RPM being used in order to solve a commitment problem between
manufacturer and retailer which arises with asymmetric information), theory does not
support a general presumptions against such restraints from a competition point of view.
Much of the empirical evidence from cross-sectional studies also supports the notion that
vertical restraints are benign and pro-competitive 3 (though again with exceptions 4).
effort is to reward the outcome – i.e. link remuneration of the platform to a metric that
proxies for success: in practice, a commission on achieved sales.
The “catch” also in this case is that if people search on the platform but “convert” on a
channel with lower commission, there is no remuneration for the “sales effort” and no
reward for the platform’s investment in quality. If consumers use the search facility but
do not “convert” the search into a booking, other channels will be free riding on the
platform. Incompleteness of contracting combined with the structure of prices which is
optimal for the platform generate clear efficiency reasons for vertical restraints in these
environments.
These are considerations specific to an online multi-sided environment that can create
new conflicts between platforms, consumers and brand owners. If control over standards
remains with the marketplace, and the brand owner as “customer” of the platform has no
control over standards but must take them as given, it may legitimately decide to retain
some control over its distribution by being selective about the sites its resellers are
allowed to use. This may well take the form of an outright restriction on resellers’ use of
certain marketplaces, or restrictions on the type of functionalities that resellers may
allowed to use on marketplaces. For instance, if a platform commits to a particular way of
doing comparisons between products, order fulfilment, payment etc., by setting
standardised terms that cannot be individually negotiated, the brand owner may not want
to be seen to be trading in that particular way and engage in the selective use of
marketplaces, and “platform exclusion” decisions.
Yet we have seen that agencies tend to regard these decisions as a form of anticompetitive
“discrimination” by brand owners across outlets, intended to prevent price competition
and “keep prices high” for their products – and as such, they tend to be seen as “per se”
anticompetitive and “by object” restrictions. An example is the ASICS case in Germany,
where it was deemed anticompetitive for ASICS to restrict resellers’ use of price
comparison websites, and the use of logo/brand on third party platforms. It is always hard
to understand in coherent economic terms why a manufacturer would want to restrict
competition in the distribution of its product – as in traditional environments, the brand
benefits from the distribution of its product being as competitive as possible. The notion
that the motive must be anticompetitive does not seem justified in light of what we know,
and of the efficiency reasons for restricting a channel of distribution.
multiple parties interacting “upstream” and “downstream”, not that there is something
inherently different arising from multi-sidedness. Indeed it is not clear that
multi-sidedness makes any essential difference to the analysis. What multi-sidedness
implies (again) is that there are search services offered for free on one side of the
platform, on which others can free ride; and these free-riding opportunities mean the
platform cannot get a return from direct search customers and may well seek to introduce
restrictions on prices for the same product sold on other channels. However, we would
have essentially the same issues also with an agency model with in which an “upstream
firm” was selling a product on a platform and setting the final price, while at the same
time selling the product as an input to the platform at a price either negotiated or set by
the platform. In practice there is a set of “vertically dis-integrated” offers (the platform
which provides the “downstream” booking service only) competing with “vertically
integrated” offers (the seller which provides the “upstream” product and competes for
sales with the dis-integrated seller through multiple channels), and this creates additional
trade-offs, but the analysis of competitive effects does not require a change in our
analytical tools: it still requires us to gauge the extent to which simultaneously selling a
product through multiple independent platforms, and through the seller’s own integrated
channels, may give rise to foreclosure incentives upstream and downstream between the
platform and its pay customers.
The area where most progress needs to be made – and where tools need to be sharpened –
is the testing of the efficiency motivations for the contractual restrictions that we see.
Competition authorities and the courts have rarely if ever accepted contractual
incompleteness as a motivation for vertical restraints in more traditional environment, and
have not engaged with the task of properly understanding organisational structures and
business models: theories that explain organisational structures and their efficiency
properties are typically dismissed, and this problem has carried over entirely to the multi-
sided environment.
Thus the case against hotel booking platforms has been strongly motivated by a prior that
MFNs/Best Price Clauses imposed by platforms on their suppliers (hotels) to ensure they
were not selling rooms at a discount on other platforms and their own sites were no more
than a form of RPM, intended to increase prices and deter the entry of cheaper platforms.
A number of analyses have been more subtle, 7 but the prevailing view was that booking
platforms somehow “squeeze themselves” between the customer and the hotel, and there
is nothing wrong with customers searching on booking sites and then booking with the
hotels separately (“information on the internet is by its nature free”). The argument that
these restraints are efficient because by increasing the “conversion” of search on the
platform into sales they increase the incentives to invest in search (only search that leads
to booking is rewarded) has been fundamentally set aside. The efficiency motivation have
been systematically “disbelieved” by agencies (“I understand the argument for
efficiencies, I just don’t believe it”). In the more elaborate version of the argument, the
answer has been that there can be no concern about the effect of free riding on incentives
to invest because the investment of the platform are not “specific” to a particular hotel –
thus if there is free riding on the part of a particular hotel, this does not undermine the
incentive of the platform to invest overall.
But it is simply incorrect to dismiss efficiency motivations on these grounds: a website
which involved a material investment to design and launch is not protected from free
riding concerns just because the investment in the technology was not “seller specific”.
Of course it is the case that for free riding to undermine the incentive to invest it must be
the case that the investment is “relationship specific”; but to conflate this with “seller
specific” is a mischaracterisation of the economic insight. What platforms are doing here
is creating a public good for everyone who searches the website, and the investment is
still specific in that sense. When pursuing arguments sourced from the economic literature
we need to be careful to capture their true meaning and substance. What happens on the
platform is a specific investment, because it is all about creating a public good for the
other side of the market.
Similarly, for selective distribution the key is that given the uncertainties of online
selling, brand owners want more residual control rights over decisions about how to retail
their products, and want authority ex post to deal with issues that might arise in an
uncertain environment. It is simply not possible for a competition agency to assess
whether the parties can write a complete a contract or not. A typical reaction is “we don’t
accept your efficiency arguments, because the incentive problem can be solved through a
two part tariff”. But this is incorrect: two-part tariffs can only resolve some types of
vertical inefficiencies, and by no means all. 8
In practice very little progress has been made in developing an understanding of how to
assess the credibility and significance of efficiency motivations for contractual
restrictions of the type considered here. In the hotel booking case the economists advising
the platforms sought to run various experiments to substantiate the claim that in the
absence of the clauses, conversion through the booking platform would decline – which
was at least the first “building block” in an analysis of the potential for free riding
concerns undermining the incentive to invest in the platform in the first place. One
“natural experiment” was made possible by the fact that in Germany the HRS platform
had been banned from using MFNs altogether, and this provided an opportunity for
studying whether this had a material effect in terms of inducing lower conversion rates on
the HRS platform (i.e. fewer bookings relative to searches) once the ban came into force.
Evidence was also collected from platform search and booking data in other countries to
assess whether conversion rates varied with the degree of “price dispersion” – i.e. the
extent to which consumers were more likely to make a booking (“convert” their search)
on the platform when prices for hotel rooms were more uniform, and less likely to do so
when they were faced with greater dispersion of room rates. The experiment was not
“clean” in the sense that MFNs were in place, and therefore the degree of price dispersion
which was observed was only reflecting “lack of adherence” to MFNs. However an
interesting claim was the finding that where price dispersion was higher (i.e. MFN were
not being adhered to), there was a material decline in the probability that customers
would book through the platform, even though they continued to use the search
functionality. This type of evidence gained little traction and was given little weight in the
case.
6. Focus on short term price effects vs dynamic effects has long term costs
efficiency benefits are achieved. Separating the analysis into stages, placing the burden of
proof on the parties to prove efficiencies and somehow show they offset the restriction, is
not how we should proceed. Weighing anticompetitive effects against efficiency benefits
is not how our economic theories work. Different is the case of mergers, where we
evaluate the change in incentives while leaving the cost structure unchanged. But
contractual vertical restraints change the incentives to compete in price vs. the quality
dimension, and it is just not possible to separate a price increase motive from an
efficiency motive. The burden of proof on efficiencies is simply impossible to meet.
And indeed, because it is seen as all too difficult to make this “balancing” assessment of
restrictions and efficiencies, there has been a major lurch back towards the use of the
hardcore “object box”, so we do not have to worry about efficiencies at all. 9 In an online
environment all forms of internet retailing are labelled “passive sales”, so that every
restriction one might want to adopt for efficiency reasons can be labelled a violation of
object.
A second factor may be the bias of Industrial Organisation for looking too much at the
details of specific models, and less at the bigger qualitative questions that matter for
policy. Related to this, much of our competition policy advice is that we should get prices
as close as possible to marginal cost at all times. We worry about restrictions increasing
prices in the short term, though we know that higher prices can be good because they
signal profitable market niches and high consumer demand, and direct investments and
entry to where the highest marginal values are. We hamper this market process of
eliciting information about demand with too much focus on static short-run price
competition and too little on market dynamics, yet we have not been particularly helpful
in suggesting evidentiary standards that are implementable in practice. We live by the
legacy of “example economics”: “there is a paper that shows that this practice can be
problematic, we have an intuition things may go this way, so it is better to be prudent”.
If we downplay efficiency arguments, and require companies to show “objective
justification” for a business practice, we are adopting the opposite of a model in which
innovation is driven by experimentation. In traditional industries in which business
format did not change as much with the product, that might not have been that much of a
problem, but with internet retailing and platform markets the freedom to experiment in
sales strategy and business format is much more central. A direct implication is we are
seeing business format change under significant strain in Europe, and a return to vertical
integration into distribution: brands selling increasingly through flagship stores or by
renting shop-in-shop modules in department stores. The purpose is to regain control over
the vertical chain, in an environment in which there is perceived great uncertainty about
what is allowed and what is not in terms of online distribution. Would we have seen so
much vertical integration if firms could control their vertical sales channels via contracts?
Probably not. Firms self-provide distribution services to regain control over their product,
but this development is induced as a reaction to concerns about enforcement in this area,
and may lead to foregoing or restricting forms of innovation that could take place online.
We are in danger of undermining the rate of innovation in this area by yielding to firms
that would like to see competition authorities shift rents to them,
There is no unique test that can be implemented to assess the potential anticompetitive
effects of vertical contractual restraints in multi-sided environments. And because the
formal analysis can be complicated by effects going in different directions, it will be
important to remain focused on “first-order” economic effects. The first priority in our
view is to establish a framework for conducting the analysis that makes economic sense
and reflects the insights we have from the economics of vertical restraints, multi-
sidedness, online and network effects. We sketch below a possible roadmap.
competitive constraints properly leads to market boundaries being too narrow and market
power being overstated for purposes of assessing vertical contracting practices.
First, almost all positive-price customers in multi-sided markets multi-home. They tend to
be sellers to end-customers and will use all distribution channels that can add to their
margins. The key is that with multihoming all distribution channels tend to be substitutes
to some extent. That multiple distribution channels tend to be used in equilibrium is not a
sign that they are complements or independent (which tends to be the standard view).
Selling through more channels is pro-competitive in itself because it reduces the marginal
contribution of each distribution channel to a firm. But since no supply channel can
extract more than its marginal contribution to downstream profits, with more distribution
channels the prices paid by firms for each particular channel will be lower. This effect is
more pronounced, the greater the degree of multi-homing and the greater the transparency
of different offers to the end customers.
Second, and closely related, digitalisation has made it much easier to offer services
through a variety of business formats, both vertically integrated and dis-integrated. The
fundamental innovation of platform markets is precisely that complementary components
to services can “plug-in” to already existing services, facilitating intermediation on any
type of service. Online retailing integrates multiple functions such as product information,
product search and matching/choice, financial transacting, and physical transportation.
These activities involve different costs, and customers typically demand different
mixtures of these activities which are offered in all kinds of combinations, with different
degrees of vertical integration, and often with a mixture of digital and traditional markets.
Thus a firm can make a sale as a result of consumers searching on a price comparison
site, and then clicking through to the firm’s website, or through search and purchasing on
a booking/sale platform; these are substitutes in the economic sense, even though in the
first case the seller pays the price comparison site for the click through, in the second it
pays the platform for click through and fulfilment. Similarly, a brand can reach customers
through a click advertisement on Google Products but also as a result of the customer
searching for the product on Amazon and buying there. In the case of a direct purchase
from the manufacturer the order may be fulfilled through an external contract with UPS,
in the Amazon case through Amazon fulfilment, but in principle these are substitutable
packages.
The analysis ought to start from a description of all channels through which an end-
consumer can be reached, and the departing presumption should be these are potentially
in the same market. However this is not what happens. The analysis often starts with a
description of the “experience” of different distribution channels, to conclude on that
qualitative basis that a number of them can be excluded from “the market”. The simple
argument that the “online experience” is different from the “offline experience” may well
establish product differentiation, but it is not enough to exclude substitution a priori. We
also often find that a distinction is drawn by pointing to the fact that different firms, e.g.
price comparison sites and sites that allow search and booking, do not offer the same
services. But this is incorrect, because what matters is the substitution between the “full
stack” including the price comparison site together with whatever financial transaction
and physical fulfilment solution they offer, and a site which provides an integrated
facility for product search, selection and financial transaction.
Several practical steps should be followed for a proper analysis:
• First, all channels have to be identified through which end-consumers can be
reached by the positive-price customer of the multi-sided platform. All such
framework must be specified before starting the evidence gathering, so that it can
discipline the interpretation of evidence and avoid the conjectural approaches that are
currently most common in complex cases.
writing complete contracts on all aspects of the actions of the contract partners. These
business reasons need to be seriously engaged with by competition agencies, but currently
they are not. Efficiency claims are routinely dismissed with reference to some contract
that the firm could theoretically write that would eliminate the problem: for instance, that
a contract could in fact be written that conditioned on the very variable with respect to
which the contract is incomplete; or that all contracting problems could be solved with
two-part tariffs.
These claims are not economically justified (just as the claim that “free riding problems
do not exist”). There is ample evidence from everyday life and from the economic
literature that (a) free riding is particularly pervasive in digital environments because its
costs have declined, (b) the predictions of incomplete contracts theory explain shifts in
ownership and contract structure in a multitude of markets. Furthermore, the theoretical
literature makes clear that two-part tariffs solve incentive problems only in a non-generic
set of cases, and they fail whenever firms in a contracting environment are not risk
neutral. A first step in the analysis of efficiencies of vertical restraints should be for these
simple principles to be acknowledged by competition authorities.
Second, we should not be asking firms to prove a negative: that there is never any other
possible contract that could possibly have the same effect on resolving the contracting
problem, but may not have potentially some anti-competitive effect. It is obvious that this
will not be possible. Firms are required by precedent to show “objective justification” for
a practice, but if what is “objective” is an entirely subjective assessment by case teams
with strong priors, there is risk for firms whose vertical restraints have in practice no
anticompetitive effects.
The basic issue is that the standards for proving efficiencies have been made impossibly
high, while the standards for proving infringements are much lower. Of course it is
entirely reasonable that to prove an infringement one does not have to show actual effects
in many cases. This would be an impossible standard and would end effective
enforcement. Quite reasonably the standard has been set to “likely effects”, which can
then be proven by a coherent theoretical framework and evidence that it applies (or even
evidence that quite regularly in similar circumstances there have been anticompetitive
effects). As there is little evidence for strong and widespread anticompetitive effects of
vertical restraints, this is in practice a very low standard of proof – even if it is a
reasonable one. However, it is then fundamentally wrong to set an impossibly high
standard of proof for efficiency defences.
We therefore propose that efficiencies should be treated to the same standard as
anticompetitive effects:
• There should be a clear theory for why the vertical restraints have been adopted ;
• The assumptions of the theory should hold in the particular market; and
• The predictions of the theory should be more consistent with the facts of the case
than the anticompetitive theory the competition authority pursues.
In practice this would mean that the burden of proof for the efficiency defence should
depend on the strength of evidence for the theory harm. For example, if a theory of
entry-deterring effects is found to be inconsistent with the pricing behaviour of entrants,
the efficiency explanation for the behaviour should gain greater weight.
Overall, the priority for the foundation of a more effective assessment of efficiency
defences does not lie in new techniques of analysis, but in creating standards of proof
for efficiencies that can actually be met, and that are nothing more than the
equivalent to the low standards which are routinely applied to “prove” likely
anticompetitive effect.
8. Conclusions
Economic analysis has failed to inform a rational policy towards vertical restraints in
Europe, and this basic failure is carrying over to multi-sided environments. We have
argued in this paper that the appropriate response is not to call for a separate toolkit for
the analysis of vertical restraints in multi-sided markets. On the contrary, most multi-
sided markets can be reinterpreted for the purposes of analysis in antitrust cases as
a standard contracting problem in vertically related markets.
Most progress can therefore be made if we are able to adopt a simple structure of analysis
that should be in principle familiar, but is not rigorously applied even in standard vertical
cases. Our recommendations for the analytical framework is therefore not to focus on
adapting techniques, but on the approach to the analysis. Just some improvements in
approach to market definition and a systematic use of the assumptions and implications of
theories of harm would lead to a much more reliable analysis of vertical restraints cases.
The call for new techniques may in part be a symptom of Industrial Organisation looking
too much at the details of specific models and less at the bigger qualitative questions,
which features of markets are important when considering policy intervention. It is not
enough to tease out and try to trade off every conceivable effect of best price clauses, for
instance, if these insights are not embedded in an investigative procedure that allows
relevant and irrelevant theories to be distinguished from each other. It is unimportant that
“there are models” showing anticompetitive effects, the key is whether such models need
assumptions that tightly map into the market circumstances of the case. This is something
we do not have enough clarity about (and discipline) in practice. The priority is to make
sure that investigations put a process into place that makes the applicability of a specific
theory directly testable, and makes this a stringent requirement – rather than relying on
general “findings” and theoretical result in the literature to justify a prior.
Notes
1 See also Kühn, K.U., “Economic Deficits in the Competition Policy Analysis of e-
commerce: is the Current Enforcement Practice Justified from An Economic Perspective?”,
in Competition Law and Policy Debate, Volume 2, Issue 2, June 2016.
2 For a broad overview see among others the classic Rey, P. and T. Vergé, 2008, “Economics
of Vertical Restraints”, in P. Buccirossi, Ed, Handbook of Antitrust Economics, MIT Press,
pp. 354 – 390.
3 See, Lafontaine, F. and M. Slade, 2008, “Exclusive Contracts and Vertical Restraints:
Empirical Evidence and Public Policy”, in P. Buccirossi, Ed, Handbook of Antitrust
Economics, MIT Press, pp. 391 – 414.
4 MacKay, A., and Smith, D.A., 2016 “The Empirical Effects of Minimum Resale Price
Maintenance”, Kilts Center for Marketing at Chicago Booth – Nielsen Dataset Paper Series
2-006, at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2513533.
5 For an analysis of the importance of choosing the right “business model, see Rochet, J.C. and
J. Tirole, 2003, “Platform Competition in Two-Sided Markets”, Journal of the European
Economic Association, 9:4, pp. 990 – 1029.
6 See inter alia Johnson, Justin P., The Agency Model and MFN Clauses (January 25, 2017).
Available at SSRN: https://ssrn.com/abstract=2217849 or
http://dx.doi.org/10.2139/ssrn.2217849 and Boik, A.K. and K.S. Corts, 2013, “The Effects of
Platform MFNs on Competition and Entry, Research Paper, available at
https://www.researchgate.net/profile/Andre_Boik/publication/305220937_The_Effects_of_P
latform_MFNs_on_Competition_and_Entry/links/5785173408ae3949cf5384ee.pdf
7 See inter alia Fletcher, A., and Hviid, M., Retail Price MFNs: Are they RPM ‘at its Worst’?,
ESRC Centre for Competition Policy, University of East Anglia, April 2014.
8 See for example Rey-Vergé (2008). Ibid.
9 See e.g. Pierre Fabre judgment ECJ 2011, which deems selective distribution an object
restriction in the absence of “objective justification’.
www.oecd.org/daf/competition